<?xml version="1.0" encoding="utf-8"?>
<rss version="2.0" xmlns:media="http://search.yahoo.com/mrss/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:lab="https://labradorcms.com/ns/rss">
<channel>
    <title>www.theregister.com - Articles</title>
    <link>https://www.theregister.com</link>
    <description>Articles from www.theregister.com</description>

    <item>
        <guid isPermaLink="true">https://www.theregister.com/a/5253818</guid>
        <link>https://www.theregister.com/offbeat/2026/06/10/raspberry-pi-project-gives-media-libraries-a-vcr-style-makeover/5253818</link>
        <pubDate>Wed, 10 Jun 2026 22:56:06 +0200</pubDate>
        <title>Blockbuster new Raspberry Pi project turns any screen into old-school VCR</title>
        <description><![CDATA[ Who needs fancy menus and high definition? 240-MP will play your media files like it's 1999 ]]></description>
        <category>offbeat</category>
                <lab:kicker><![CDATA[ offbeat ]]></lab:kicker>
                <content:encoded><![CDATA[ I love Star Trek so much. I’ve watched most Trek series multiple times over the decades, and was shocked when, on my most recent watch of The Next Generation, I noticed something: High definition upscaling makes the show look way worse. Old-school 4:3 CRT television screens with their low resolution hid a lot of stuff, like tape on the Enterprise set doors that hid whatever names were stenciled on them for prior episodes, which are glaringly present on modern editions of the show. I’ve always been on the lookout for a way to capture the classic Trek feeling, and one … ahem … enterprising developer has done just that. Anthony Caccese, a principal product lead for enterprise platforms at Oak Ridge National Laboratory by day and a Raspberry Pi tinkerer by night, recently published an open-source project called 240-MP on GitHub. It’s a simple concept: Text-based menus that look like an old-school VCR interface, but with modern functionality and, most importantly, the ability to play local media files and Plex libraries on an old-school CRT TV. 240-MP runs on a Raspberry Pi, is based on the command-line media player MPV, and can play local files (either on the Pi itself, a USB drive, an external hard disk, or even a network share) or media from a Plex server, as Caccese built modules for both local and Plex-based playback. If you don’t happen to have an old CRT TV or monitor lying around, or the necessary Pi-compatible composite cable to connect your SBC to said TV, 240-MP will also work with a modern screen and an HDMI connection, too. One note on the composite vs. HDMI option, as noted in the setup instructions: You will need to update the config.txt file to support one or the other, so have your output chosen ahead of time. Once the system is installed, you can navigate around 240-MP with either a remote control or a keyboard, where you’ll see text menus for navigating around to different folders, choosing episodes or playlists, switching audio and subtitle tracks, looping playback, and the like. It might look like an old-school VCR interface, but with a lot more capabilities. Caccese has only tested 240-MP on a Raspberry Pi 4B, 3B+, and 3B, noting that he’s not sure it’ll work on other devices and has no plans to test other hardware, either. What will be coming in the future, Caccese said in an accompanying YouTube video, is modules to support other media playback software, like Jellyfin (a popular Plex alternative in light of that massive price hike), and RetroArch, a frontend for emulators designed to play old-school video games. “Please feel free to fork this repo, update any aspects and tailor things to your own use case; that's why the source is fully open and available,” Caccese noted on GitHub. Now if I could only find a working CRT TV to pair with my old Raspberry Pi, I could go on a hardcore 90s nostalgia trip and feel just like I did watching VHS tapes of Star Trek episodes I recorded from the TV when I was a kid. After all, streaming high-def remasters just isn’t the same. ® ]]></content:encoded>
                <enclosure url="https://image.theregister.com/?imageId=5253841&amp;width=800" type="image/jpeg" />
                <media:thumbnail url="https://image.theregister.com/?imageId=5253841&amp;width=800" />
            </item>
    <item>
        <guid isPermaLink="true">https://www.theregister.com/a/5252373</guid>
        <link>https://www.theregister.com/software/2026/06/08/canonical-sends-ubuntu-into-the-ai-agent-era/5252373</link>
        <pubDate>Mon, 08 Jun 2026 18:34:02 +0200</pubDate>
        <title>Canonical sends Ubuntu into the AI agent era</title>
        <description><![CDATA[ Sandboxed LLM dev environments lead the show, but accessibility may be the real prize ]]></description>
        <category>software</category>
                <lab:kicker><![CDATA[ software ]]></lab:kicker>
                <content:encoded><![CDATA[ UBUNTU SUMMIT Canonical is still experimenting with the format of the Ubuntu Summit series of free conferences, and its most recent instance, the 26.04 edition, was a primarily online event. There was a small in-person invited audience, which by our informal estimate was about half the size of the one at last October's edition. The event opened with a keynote from Canonical founder Mark Shuttleworth, and his opening sentence set the tone for much of what would follow: The agentic revolution will touch every aspect of human endeavor. We take that to mean the use of LLM "agents" to develop software, translate between human languages and from speech to text, and so on. For all that this vulture might personally dispute just how revolutionary this is, there were some 21 full-length talks over the two days of the summit, and about half of them were about AI, or at least touched upon the subject. Shuttleworth's keynote also contained the biggest Canonical product announcement of the event: the new Workshop sandboxed LLM development environments (at the 20-minute mark in the video above). Workshop uses Canonical's LXD "containervisor" and snap packaging to make it easy to install and run LLM agents, while keeping them isolated in sandboxes so that they can only access specific limited resources in that user's home directory. For instance, they can access the machine's GPUs and nominated local files, while being walled off from personal data such as stored credentials. As Shuttleworth put it: You can run random code, from the internet, on your laptop, without handing it root. Canonical also announced Workshop online the same day, with a collection of documentation already available, including a tutorial. Workshop is an open source project with the source code on GitHub. Later that day, engineering manager Dmitry Lyfar gave a talk on the new tool, titled Introducing Workshop. Shuttleworth's keynote was followed by another by VP of engineering Jon Seager. As we reported last month in our article on AI integration into Ubuntu and Fedora, Seager recently published a blog post about the company's AI intentions. In his keynote, Seager said that this post had been "SEO'd to death," but he too devoted a substantial part of his talk to AI, saying: Ubuntu can't be in the conversation about AI and open source unless it has a position and a stake. Seager also spelled out some of what this will mean, from small feature improvements such as improving auto-focus in webcams and making power management more intelligent, to more significant features. He called out accessibility as a key area for investment and improvement. He said that "existing Linux screen readers suck" – harsh, but not entirely unfair – and that there is "so much room for improvement" in that area. He continued that the plan is "to enable speech-to-text everywhere in the desktop," but said "AI is transformative for people with disabilities" and that the company soon hopes to preview the "first AI-powered context-aware desktop features." In case, this sounds niche or unimportant, it really isn't. Speech-to-text is a vital tool for people with physical impairments that make typing difficult. This vulture has written at length about the importance of keyboard user interface design, as well as about how few Linux desktops fully and correctly implement it, leaving Apple with a significant edge in this area. As it happens, this author is a keyboard-intensive user with relatively poor eyesight, so this matters to us. Register accessibility columnist Colin Hughes has written about the importance of speech-to-text UI. For now, Linux's usability in this area is much poorer, and as Wayland displaces X11 from the big-name desktops, it's about to get a lot worse, as the recent blog post from "nocoffei" describes: My Accessibility Stack and the future on Wayland. "nocoffei" links to the same series of blog posts by TapType developer Aaron Hewitt that we did back in March, under the collective title "I Want to Love Linux. It Doesn't Love Me Back." We recommend them again:  Built for Control, But Not for People: Linux is already broken before you even start  The Audio Stack Is a Crime Scene: You can't hear anything – and it's not your fault  Interlude – A Thank You, Where It's Due: This is what it looks like when people care  In part 4, he takes a surprising new direction: Wayland Is Growing Up, And Now We Don't Have a Choice: The future is Wayland. Let's make sure we're invited. In that, he reports on significant strides in keyboard-driven accessibility for blind users with GNOME on Wayland – but as nocoffei's post spells out, that is no help to those who can see fine but can't type. If the integration of AI into Ubuntu can address this or improve on the current situation, that will go further toward ameliorating this vulture's deep skepticism about the viability of LLMs than anything else. Bootnotes Canonical invited The Register to attend the Summit in person, and paid for our travel and accommodation during the event. Indeed, if you look closely at the 30-second mark of the highlight reel – it's only 50 seconds long, so it's not too arduous – you can see The Reg FOSS desk's hands typing away industriously into Logseq. Back in April 2023, severe injuries from a road traffic accident very nearly cost the author one of those hands, which is part of the reason for his interest in accessibility tools – as well as the reason for the MacBook Air visible in that video, into which his articles were dictated for the next few months. On the subject of AI integration into desktop distros, it's interesting to note that since we wrote about AI tooling in Fedora a month ago, there has been considerable pushback from the user community, and two committee members have changed their votes to oppose it. ® ]]></content:encoded>
                <enclosure url="https://image.theregister.com/?imageId=5252431&amp;width=800" type="image/jpeg" />
                <media:thumbnail url="https://image.theregister.com/?imageId=5252431&amp;width=800" />
            </item>
    <item>
        <guid isPermaLink="true">https://www.theregister.com/a/5245952</guid>
        <link>https://www.theregister.com/software/2026/06/08/consultant-mistakenly-deleted-a-ton-of-data-but-reported-it-as-a-bug/5245952</link>
        <pubDate>Mon, 08 Jun 2026 08:30:00 +0200</pubDate>
        <title>Consultant mistakenly deleted a ton of data – but reported it as a bug</title>
        <description><![CDATA[ And he got away with it too! ]]></description>
        <category>software</category>
                <lab:kicker><![CDATA[ Software ]]></lab:kicker>
                <dc:modified>Tue, 02 Jun 2026 11:00:41 +0000</dc:modified>
                <content:encoded><![CDATA[ WHO, ME? Is showing up for work every Monday a mistake? While you ponder that question, dive into a new installment of Who, Me? – The Register's weekly column that shares readers' stories of escaping their errors. This week, meet a reader we'll Regomize as "Evan," who wrote to us from the side of a pool while his kids had a swimming lesson! "I work in test automation as a consultant and for one client I had to record test evidence as video," he explained, adding that the client's test management tool stored the vids. The resulting files weren't individually large, but by the time Evan had recorded 600 of them, managing all those files was starting to get a bit cumbersome. "Removing them manually was far too slow and wasn't feasible," he wrote. So he wrote a script to clean things up all at once. "Obviously this data was important, and I'm not reckless," Evan wrote. He therefore carefully debugged the script using breakpoints. "I stepped through every line, I checked all values, and I could see everything was right. Then I let the code try to delete the one file I was watching." The script deleted that file. And everything else in the container that the test tool used to store videos and plenty of other data. Did we mention this happened in the middle of a project, meaning Evan's action was profoundly unwelcome? Evan reckoned he was probably at fault, but decided not to confess to his client and instead informed them about the data loss and logged a support ticket. The client therefore assumed this incident was an accident and was cool about it. After a week of back-and-forth with support, Evan got good news. His client's support team was able to restore the data from a backup and could not find a reason for the incident. And then came even better news. "They took all ownership of the fault," Evan admitted. "They were very apologetic and said one of their SaaS scripts had gone haywire and deleted the content." Evan therefore escaped blame and carried on consulting – and is clearly doing well enough to pay for multiple kids to have swimming lessons! Have you successfully escaped blame for an error? If so, click here to send an email to Who, Me? It would be a mistake not to share your story so we can present it to your fellow readers. ® ]]></content:encoded>
                <enclosure url="https://image.theregister.com/?imageId=250181&amp;width=800" type="image/jpeg" />
                <media:thumbnail url="https://image.theregister.com/?imageId=250181&amp;width=800" />
            </item>
    <item>
        <guid isPermaLink="true">https://www.theregister.com/a/5250718</guid>
        <link>https://www.theregister.com/ai-and-ml/2026/06/03/no-longer-just-a-copilot-microsofts-ai-wants-to-take-the-wheel/5250718</link>
        <pubDate>Wed, 03 Jun 2026 18:49:15 +0200</pubDate>
        <title>No longer just a Copilot, Microsoft's AI wants to take the wheel</title>
        <description><![CDATA[ Always-on agent promises to keep work moving, provided you trust it with practically everything ]]></description>
        <category>ai and ml</category>
                <lab:kicker><![CDATA[ AI AND ML ]]></lab:kicker>
                <dc:modified>Wed, 03 Jun 2026 16:49:45 +0000</dc:modified>
                <content:encoded><![CDATA[ Move over, Copilot: Microsoft is introducing a new category of agentic AI called "Autopilot," starting with Scout, its first agent. And it doesn't take much guessing to understand how Microsoft expects these things to operate: By constantly watching your every move and taking action in the background to ostensibly streamline your workday. Microsoft announced Autopilot, and the first Autopilot agent, Scout, at Microsoft Build on Tuesday, describing it and other future Autopilots as “always-on agents that work autonomously,” stay active in the background to “understand how work gets done across your apps and systems,” and can “take action without needing to be prompted each time.” Scout, for example, can be interacted with in Teams when one feels the need, but outside of instances when users need to query it directly, it’s always there. “It operates across cloud, desktop, and web, connecting to Teams, Outlook, OneDrive, and SharePoint, and to the data that powers your day, including chats, email, calendar, and contacts,” Omar Shahine, corporate VP of Microsoft Scout, wrote in the announcement. Autopilot agents supposedly have their own identities, according to Shahine, and are able to act autonomously within the constraints organizations set on their activities (access controls can be set by organizations). Per Shahine, letting Autopilots operate on autopilot “creates a more durable way to keep work in motion so it continues even when your attention is elsewhere.” Say, for example, you need to schedule a meeting: Scout can handle scheduling on its own while accounting for time zones; it can flag meetings it considers particularly important for its users and generate materials it believes users need to prepare before the scheduled time. Scout can also identify looming deadlines and block off time on a user’s calendar so that they can work on a particular project, “spot risks, like stalled decisions,” and basically act like a work nanny that schedules your day by being hyper-aware of every single little thing that needs to get done. Hopefully, your new Microsoft nannybot is more reliable than its Copilot predecessor, whose outputs Microsoft itself warns may not always be accurate. Get ready for a Claw-shaped hole in your environment “Microsoft Scout is built with enterprise-grade security and controls so it can be trusted in your organization from day one,” Shahine noted in the release, followed immediately by noting that it’s powered by OpenClaw, not exactly a platform with a stellar security reputation or record of not making bad decisions on behalf of users. Microsoft claims that Scout and whatever future Autopilot agents it releases are bound to an Entra identity that allows their activity within an enterprise environment to be attributed to a particular person’s Scout agent, and notes that it acts within the confines of access controls set by the organization, but it’s not clear what other protections against common AI exploits are included. As we’ve noted before, it's often surprisingly easy to manipulate AI agents into behaving in ways their operators never intended, and malicious webpages can inject prompts that trick them into leaking sensitive information; in both cases, those sorts of attacks can be launched without any direct user interaction. We asked Microsoft for more details on the security aspect of Autopilots and Scout, but didn’t hear back before the deadline. It’s also worth noting that Microsoft Scout is in very limited access, with only a “select group of customers” getting access to the preview, along with organizations participating in the Frontier program, which grants them early access to Copilot and other Microsoft AI features. One more caveat, too: Frontier enrollees can only get access to the Scout preview if they’re GitHub Copilot subscribers. GitHub Copilot recently shifted to a usage-based billing model that has seen bills skyrocket, so expect those Microsoft bills to rise if you choose to give it a shot, too. ® ]]></content:encoded>
                <enclosure url="https://image.theregister.com/?imageId=5250787&amp;width=800" type="image/jpeg" />
                <media:thumbnail url="https://image.theregister.com/?imageId=5250787&amp;width=800" />
            </item>
    <item>
        <guid isPermaLink="true">https://www.theregister.com/a/5250150</guid>
        <link>https://www.theregister.com/software/2026/06/03/uk-lawmakers-call-on-government-to-ditch-palantir-nhs-contract/5250150</link>
        <pubDate>Wed, 03 Jun 2026 11:01:00 +0200</pubDate>
        <title>UK lawmakers call on government to ditch Palantir NHS contract</title>
        <description><![CDATA[ Lock-in to a small number of suppliers holding up digital government plans, committee says ]]></description>
        <category>software</category>
                <lab:kicker><![CDATA[ Public Sector ]]></lab:kicker>
                <dc:modified>Tue, 02 Jun 2026 15:37:17 +0000</dc:modified>
                <content:encoded><![CDATA[ MPs have told the government to cut its ties with Palantir, and end the US spy-tech firm's controversial involvement in the National Health Service's Federated Data Platform. Warning against vendor lock-in across government, the House of Common science and technology committee said it was most concerned about Palantir, which had secured central roles in health and defense systems. “Palantir should not have a such a significant role in the UK public sector… it is far from the only company capable of providing the data analysis ‘middleware’ required by public bodies,” the report from the Science Innovation and Technology Committee said. The report notes concerns about Palantir’s origins as a company getting a foothold in government with security, immigration services, and defense contracts. It also describes the political musings of co-founder Peter Thiel and CEO Alex Karp. However, it added: “Our view that Palantir’s increasing presence across the public sector represents an unacceptable point of weakness is not ideologically motivated or driven by concerns about the quality of their products. The government should retain the ability to pick and choose individual suppliers and safeguard against the risk of vendor lock-in and debilitating dependencies, particularly in areas of critical national importance such as healthcare and national security infrastructure.” Palantir won the £330 million Federated Data Platform (FDP) contract in November 2023 after a procurement process, which NHS England, the soon-to-be-defunct health quango, maintained was open and fair. The award followed £60 million in Covid-era NHS contracts awarded without competition. The committee recommended that the government use the February 2027 break clause in the FDP contract and either “develop an in-house replacement or seek an alternative developed by UK-owned and UK-based providers that are more compatible with UK values, and do not pursue either technical or contractual dependencies.” The Science Innovation and Technology Committee said dependency on a small number of suppliers, locked into repeated government contracts, was one of the factors holding up delivery of the broader vision for digital government. The others included over-reliance on legacy systems, the problem of digital sovereignty and over-hype by senior politicians and industry figures. Dame Chi Onwurah MP, Committee chair, said: “We welcome the government’s intentions to make the UK a ‘truly digital state,’ but it’s not clear how this will be delivered. Without a detailed and measurable plan, it risks falling short – but there’s still time to put this right. “A critical part of this transformation should include reducing the UK’s dependence on a small number of big US tech companies like Palantir. Vendor lock-in isn’t inevitable, and the current position leaves us seriously exposed. The UK can and should be aiming for technology sovereignty in critical parts of our public sector and supporting domestic alternatives through smarter procurement,” she said. The committee said the government needed to get its approach right before embarking on ambitious projects such as the digital ID scheme, expected to roll out by the end of the current Parliament. Without modernized digital infrastructure, digital ID will struggle to succeed, and to keep citizens’ data secure. “Only once the foundations of the UK’s digital infrastructure are secure, and public trust has been gained, should the government proceed with its planned digital ID. The success or failure of this project will be a defining test of its wider digital transformation ambitions," Onwurah said. ® ]]></content:encoded>
                <enclosure url="https://image.theregister.com/?imageId=228315&amp;width=800" type="image/jpeg" />
                <media:thumbnail url="https://image.theregister.com/?imageId=228315&amp;width=800" />
            </item>
    <item>
        <guid isPermaLink="true">https://www.theregister.com/a/5249753</guid>
        <link>https://www.theregister.com/ai-and-ml/2026/06/01/anthropic-now-atop-the-ai-bubble-files-for-its-ipo/5249753</link>
        <pubDate>Mon, 01 Jun 2026 20:12:22 +0200</pubDate>
        <title>Anthropic, now atop the AI bubble, files for its IPO</title>
        <description><![CDATA[ First it tops OpenAI's valuation, then it beats Altman to the IPO punch ]]></description>
        <category>ai and ml</category>
                <lab:kicker><![CDATA[ AI and ML ]]></lab:kicker>
                <dc:modified>Wed, 03 Jun 2026 15:18:09 +0000</dc:modified>
                <content:encoded><![CDATA[ Anthropic has beaten OpenAI to the IPO punch, just days after its latest private funding round eclipsed its top rival’s valuation, setting up a showdown that could pump more air into - or finally pop - the AI bubble. Anthropic said Monday in a press release that it had filed a confidential S-1 form with the US Securities and Exchange Commission, setting the stage for an eventual initial public offering of shares in the world’s (currently) most valuable AI startup. Given it’s a confidential filing, Anthropic shared little information and declined to answer questions on the matter. “The proposed initial public offering will depend on market conditions and other factors,” Anthropic said in the announcement. “The number of shares to be offered and the price have not yet been set.” Those market conditions could look like anything by the time the SEC finishes reviewing the filing and it’s made available to the public for scrutiny, but the winds inflating the AI bubble are currently blowing in Anthropic’s favor. The company reported last week that it completed a $65 billion series H funding round, pushing its post-money valuation to $965 billion, sending it rocketing past OpenAI’s most recently-reported valuation of $852 billion, which at the time was the highest-ever valuation of a pre-IPO tech company. With Anthropic now on the top of the heap, it’s a perfect time to file an IPO prospectus before Altman and company steal the lead back. That said, Anthropic has made some clever choices that have put it atop the heap for now. Claude Code has done wonders for Anthropic’s reputation as the more useful of the pair, which is backed up by the fact that Anthropic reportedly earns more in revenue despite having a fraction of OpenAI’s user base. Of course, that doesn’t mean Anthropic’s valuation is realistic, nor that it’s actually posting a profit. According to a recent story in the Wall Street Journal, Anthropic is on the verge of reporting its first quarter of operating profit, according to people at the company who spoke anonymously. The Journal also notes that, as a private company, Anthropic isn’t required to post financials or report numbers any more realistic than what one would get with any good ass-pull – the word "profit" could exclude all sorts of expenses. Toss in an active Series H round (the WSJ piece was published prior to Anthropic announcing its recent valuation) and a looming IPO, and the realism of that profitability figure is questionable. We won’t be able to accurately assess the state of Anthropic’s finances until that IPO filing becomes public, which could end up serving as the first real look behind the shroud of fiscal secrecy that AI firms have operated behind. If Anthropic’s numbers look anything like what SpaceX’s IPO filing revealed (i.e., ridiculous valuations on top of massive losses) it could cause the AI bubble to start looking even flimsier than it already does. ® ]]></content:encoded>
                <enclosure url="https://image.theregister.com/?imageId=4093866&amp;width=800" type="image/jpeg" />
                <media:thumbnail url="https://image.theregister.com/?imageId=4093866&amp;width=800" />
            </item>
    <item>
        <guid isPermaLink="true">https://www.theregister.com/a/5248491</guid>
        <link>https://www.theregister.com/software/2026/05/30/wikipedia-editors-plot-strike-and-banner-sabotage-after-wikimedia-layoffs/5248491</link>
        <pubDate>Sat, 30 May 2026 13:15:00 +0200</pubDate>
        <title>Wikipedia editors plot strike and banner sabotage after Wikimedia layoffs</title>
        <description><![CDATA[ Foundation sparks revolt after disbanding team responsible for many community-requested fixes and moderation tools ]]></description>
        <category>software</category>
                <lab:kicker><![CDATA[ Software ]]></lab:kicker>
                <dc:modified>Fri, 29 May 2026 14:14:49 +0000</dc:modified>
                <content:encoded><![CDATA[ The Wikimedia Foundation (WMF) has sparked a revolt among Wikipedia editors after disbanding the engineering team responsible for many community-requested fixes and moderation tools. The Register was tipped off this week to growing unrest inside the Wikipedia editing community following the WMF's decision to disband its Community Tech team, the group responsible for triaging and developing editor-requested bug fixes, moderation tools, and workflow improvements through the long-running Community Wishlist process. Wikimedia's internal forums have turned into a running argument over how editors should respond. Some are calling for editing strikes, while others want volunteers to stop handling vandalism cleanup for a period of time. There have also been discussions about replacing fundraising banners with messages criticizing the layoffs. The foundation confirmed to The Register that the restructuring affected six staff roles connected to the Community Wishlist program, including engineers and a manager. It said the decision came after months of internal reviews that started last year. According to the foundation, leadership concluded that relying on a single dedicated team to process editor requests was no longer working well. "We learned from these assessments that it is rarely possible to fulfill community wishes through a single team due to the vast breadth of the software we support and the number of channels through which we receive wishes," a spokesperson for the foundation said. Under the new structure, responsibility for Community Wishlist requests will be spread across the wider Product and Technology department rather than handled by a dedicated team. The foundation said affected employees remain employed for now while being considered for other internal roles. Staff who are not placed elsewhere inside the organization will leave next month with severance packages. That explanation has gone down badly with parts of the editor community, where some contributors accuse Wikimedia leadership of becoming increasingly disconnected from the unpaid volunteers who maintain Wikipedia itself. Several editors have also questioned why an organization reporting nearly $300 million in assets in its latest annual report is restructuring an engineering team dedicated specifically to editor support. The situation has become even messier because several affected employees were reportedly involved in early unionization efforts linked to a newly created labor group called Wiki Workers United. One of the laid-off engineers created the union page on Wikimedia Meta earlier this month, fueling accusations from some editors that the restructuring amounted to union busting. The foundation denied that outright, telling The Register: "The decision to disband the Community Tech team is not in any way connected to discussions about unionizing, nor have we terminated any staff for their participation in those discussions." The WMF also stressed that no formal request for union recognition has been submitted and said it would respect the legal process if staff eventually vote to unionize. Meanwhile, editors continue to discuss protest options that could create highly visible problems for the world's largest online encyclopedia. Since much of Wikipedia's moderation infrastructure is maintained by volunteers rather than foundation employees, even a temporary pullback in anti-vandalism work could turn parts of the site into an open sewer of spam, hoaxes, and defacement. ® ]]></content:encoded>
                <enclosure url="https://image.theregister.com/?imageId=243499&amp;width=800" type="image/jpeg" />
                <media:thumbnail url="https://image.theregister.com/?imageId=243499&amp;width=800" />
            </item>
    <item>
        <guid isPermaLink="true">https://www.theregister.com/a/5248474</guid>
        <link>https://www.theregister.com/software/2026/05/29/ucla-seeks-pre-litigation-resolution-with-oracle/5248474</link>
        <pubDate>Fri, 29 May 2026 17:21:20 +0200</pubDate>
        <title>UCLA seeks pre-litigation resolution with Oracle</title>
        <description><![CDATA[ Discussion understood to concern delayed SaaS transformation project ]]></description>
        <category>software</category>
                <lab:kicker><![CDATA[ software ]]></lab:kicker>
                <dc:modified>Sun, 31 May 2026 10:35:22 +0000</dc:modified>
                <content:encoded><![CDATA[ UCLA has entered pre-litigation discussions with Oracle, one of the suppliers underpinning its finance transformation project, which has been delayed by nearly six years. The project has been on pause since August 2024 while the giant US university considers its options, including whether to continue using the vendor. The Regents of the University of California's Compliance and Audit Committee recently listed a proposed settlement with Oracle America over an alleged contract breach as an action line on its meeting agenda. UCLA Faculty Association member Dan Mitchell noted the item “likely refers to the failed Ascend 2.0 matter”, referring to the project name for the university’s finance and procurement system transformation. A spokesperson for UCLA said in a statement emailed to The Register: "UCLA does not comment on confidential pre-litigation matters or potential settlements. The university continues to evaluate the most effective path forward for financial systems modernization." A report presented to the University Regents Finance and Capital Strategies Committee in March last year said that among the “Top Issues, Risks and Challenges” the UCLA project leadership faced was “Oracle’s lack of responsiveness — particularly regarding licensing costs and support.” The report, based on data from up to December 2024, listed it among issues under ongoing evaluation. It shows the project's original $120 million budget was revised down to $98.9 million, with $13.5 million spent to date. Oracle has so far declined the opportunity to comment. The project started in April 2018 and was originally expected to end in July 2020. It is a “comprehensive business transformation initiative designed to modernize the University financial, budgetary, and research administration operations by migrating to the Oracle Cloud SaaS solution,” the document said. It also included retrofitting any systems which connected with the main finance system. The procurement module, BruinBuy Plus went live in January 2024, but the main “Oracle Financials go-live has been paused since August 2 [2024] and is undergoing program assessment,” the document said. Among the mitigation plans in the review was to “finalize the decision on whether to continue with the current vendor or explore alternatives” based on a reassessment of the tool and the provider. The university is yet to announce the results of planned work to “determine the viability of the current software provider and explore alternatives if needed.” According to a report from campus newspaper Daily Bruin, UCLA currently uses legacy financial systems software designed in the 1980s when the university’s operating budget was just 7 percent of its current size. It is a mainframe-based "ancient relic of a system" one interview said. A presentation given during the May 2024 Ascend 2.0 quarterly town hall estimated the total cost was projected to be roughly $286 million, with around $213 million already spent, the report said. ® ]]></content:encoded>
                <enclosure url="https://image.theregister.com/?imageId=5248522&amp;width=800" type="image/jpeg" />
                <media:thumbnail url="https://image.theregister.com/?imageId=5248522&amp;width=800" />
            </item>
    <item>
        <guid isPermaLink="true">https://www.theregister.com/a/5247224</guid>
        <link>https://www.theregister.com/software/2026/05/29/that-an-app-fits-on-a-floppy-is-still-a-useful-measure-in-2026/5247224</link>
        <pubDate>Fri, 29 May 2026 11:15:00 +0200</pubDate>
        <title>That an app 'Fits on a Floppy' is still a useful measure in 2026</title>
        <description><![CDATA[ In a world of mass-produced bot-slopware, small is more beautiful than ever ]]></description>
        <category>software</category>
                <lab:kicker><![CDATA[ Software ]]></lab:kicker>
                <dc:modified>Sun, 31 May 2026 10:33:05 +0000</dc:modified>
                <content:encoded><![CDATA[ If you're old enough, you might remember using floppy disks, either of the 3.5-inch or 5.25-inch variety. They didn't hold much and often you had to have many disks to install one program. Don't be misled by Fits on a Floppy's retro-tech name: it is most definitely not about 20th century data media. It's about compactness and comprehensibility. Fits on a Floppy describes itself as "a Manifesto for Small Software," and as we read it, we found ourselves nodding in agreement, right from the opening line: That is certainly the impression of this author, and it is not just us. We are irresistibly reminded of the Red Hat developer's six waves of industry BS that we recounted in February. Like any eternal verity of the computing industry, there's even an XKCD comic about it, if you needed any more persuading. XKCD's own internal citations, both about voting machines and indeed about the use of the blockchain, reinforce the message. Randall Monroe spells it out: That sounds about right. And parenthetically, anyone who says that they can improve anything with either blockchain or AI is no more to be trusted than a schoolteacher who gives no homework. A year before that Red Hat engineer talked to us about waves of industry BS, Dutch consultant Bert Hubert talked to The Reg about digital sovereignty – and he feels similarly. In 2024, he wrote A 2024 Plea for Lean Software in tribute to the great Niklaus Wirth, who passed away earlier that month. In our own obituary for Professor Wirth, we mentioned the 1995 paper that inspired Hubert: A Plea for Lean Software (this is a PDF of a scan – but we have posted a more readable plain text version here). One of the lasting effects of that paper is what is now called Wirth's Law: Sadly, that seems to have been its main impact. As a usable demonstration of his 2024 "plea," Hubert offered a working example, which he explained in Trifecta Technology. It's a web image-sharing tool, implemented in under 2000 lines of code. There's also a page for the Trifecta app itself, which comes in as a 1.7 MB compressed Docker file. (With some clever disk formatting, you could get that on a 1.4 MB floppy.) There are, as that anonymous Red Hatter observed, so many different layers of unnecessary complication and plain old marketing lies in modern IT that it is now hard to even keep track of them. One point of the Fits on a Floppy idea is that if you impose an artificial limit on project size, merely by keeping it very small, you will be forced to keep it very simple. That simplicity is the goal here, not fitting on 1980s physical media. You might react with scorn when you hear the idea that in the 2020s, anything useful could fit into under 1.5 MB. When even a leading tool to write an ISO file onto a USB key is a hundred times that size, it sounds absurd. But it really is not. The mind behind the manifesto is developer Matt Sephton, and he offers 18 tiny but useful apps that he's written to prove his point – plus a screensaver which we feel sure is an hommage to Berkeley's classic Flying Toasters screensaver. Others are still making useful single-floppy-sized apps today. We wrote about the revival of the Dillo web browser, and at last year's FOSDEM, the project lead was handing out floppies with the latest release. The whole app, on one diskette. Drew DeVault's Hare programming language is still in development, but when it reaches version 1.0, he plans to sell copies on a floppy: Another tiny modern language is the Janet Language. It's not quite so small, but its just over 2MB download could fit onto the 2.8 MB floppies that were used in later IBM PS/2 models and the NeXTstation. The real point here is about the readability and long term maintainability of compact, even minimalistic code. It's a similar point to that made in Dave Gauer's Ascetic Computing essay, which we cited and linked to when looking at OpenBSD 7.9. Small size and simplicity is what Fits on a Floppy is really talking about, not about physical media. He explicitly spells it out for the hard-of-thinking: Bert Hubert too returned to this theme when he wrote a piece On Long Term Software Development. At this year's Open Source Policy Summit, we saw some pundits pontificating that to escape the US cloud, the answer was that Europe needed its own companies running their own datacenters running Europe's own domestic cloud. This is so manifestly Getting The Wrong End Of The Stick that it put us in mind of Wolfgang Pauli's famous line: "That is not only not right; it is not even wrong." The way to escape a broken model that was a bad idea in the first place is not to make your own sovereign version. All you're doing is locking yourself in your own personal cage. The smart answer is to discard the broken model, and go back to an older, simpler model where organizations own and store their own data on their own servers. As ever, the KISS Principle is one of the best guidelines. It's Occam's Razor in reverse: the best solution is the simplest possible solution. If the problem is that you are trapped in someone else's cloud, then don't switch to another cloud and risk it happening again: get your private data out of the cloud altogether. Just Use One Big Server. Hire some grumpy old techies with grey hair (or none) to run it – there are plenty out there, but ageism keeps them out of work. At the smallest and most local end of this scale, then one useful guiding principle is to just keep the tools as small as you can possibly make them. It's an artificial limit, but that does not lessen its validity. It's not the only way. It may not even be the best way. But it's a way, a simple, clear, obvious way – and there's nothing to prevent anyone finding their own different path to radical simplicity. The PC rose to greatness running on two 360 kB floppy drives – hard disk drives only came along later. Tools like Lotus 1-2-3 redefined business management running on one 360 kB disk, with a second 360 kB data disk in drive B: – and this vulture is willing to bet that some spreadsheets built on such machines, long since converted to Microsoft Excel, are still running multinationals, and indeed nations, today. Compared to that, 1.4 MB is luxurious. ® ]]></content:encoded>
                <enclosure url="https://image.theregister.com/?imageId=5247272&amp;width=800" type="image/jpeg" />
                <media:thumbnail url="https://image.theregister.com/?imageId=5247272&amp;width=800" />
            </item>
    <item>
        <guid isPermaLink="true">https://www.theregister.com/a/5247916</guid>
        <link>https://www.theregister.com/software/2026/05/28/zig-creator-seeks-uncompromising-perfection-before-blessing-10/5247916</link>
        <pubDate>Thu, 28 May 2026 17:00:00 +0200</pubDate>
        <title>Zig creator seeks 'uncompromising perfection' before blessing 1.0</title>
        <description><![CDATA[ Andrew Kelley interview describes paying monthly for cloud-powered AI coding as an 'insane proposition' ]]></description>
        <category>software</category>
                <lab:kicker><![CDATA[ Software ]]></lab:kicker>
                <dc:modified>Thu, 28 May 2026 20:30:26 +0000</dc:modified>
                <content:encoded><![CDATA[ Andrew Kelley, inventor and BDFL (benevolent dictator for life) of the Zig programming language, was interviewed by Vitaly Bragilevsky, head of the Rust ecosystem at tool vendor JetBrains. Zig is a general-purpose language that aims to be as performant as C but with "fewer footguns," in Kelley's words. It is a niche language, at 82 on the latest RedMonk Programming Language Rankings, but well-liked by its users; last year's Stack Overflow survey has Zig as the fourth most admired language, defined as Zig developers who want to continue using the language. We found the interview disjointed, perhaps because of the way it was edited, but nevertheless it touches on many of the key questions developers face today, including AI, GitHub reliability problems, and programming language choices. Kelley describes why he created Zig, when other options including C, C++, Rust, and Go already exist. He said he set out to develop a digital audio workstation. He tried Go, but found interoperability with C libraries difficult, and said the garbage collector caused audio delays. He tried C++, or coding C-style using a C++ compiler, but found that small mistakes led to memory corruption bugs that took weeks to fix. He tried Rust but "really struggled to write code that would satisfy Rust's rules," and spent a month trying to make font rendering work. Kelley's solution was to go in a different direction and create a new programming language. Zig, he said, "does not give up any of the power that C offers, while improving the flaws and weaknesses that C has." As far as we can tell, the digital audio project remains in its early stages. The Zig project is known for its no-AI policy, set out in the code of conduct. The reason, he said, is that AI contributions are "invariably garbage" and consume code review time that the team believes is better spent on human contributions. His view on AI, though, is more nuanced than these first remarks suggest. One of his objections to AI is that it is unteachable, whereas the team values mentoring contributors so that they may later become better contributors or even part of the core team. Another negative for AI tooling is that it is non-deterministic and therefore its output always needs review, even for something as simple as refactoring the name of a function. Kelley prefers deterministic tools in which he can have full confidence. Regarding vibe coding – delegating all coding to AI – Kelley said it is interesting but he does not wish to use technology "controlled by four companies." He also said: "I'm not going to go from using my own computer and my own electricity, in order to use closed-source programming on someone else's computer through the network, that I have to pay for monthly. To me, that is an insane proposition." "I'm always hearing people say that AI code works surprisingly well. But to me, that is not the bar that I want to hold software to. The bar that I want to hold software to is uncompromising perfection." This perfectionist attitude is also evident in the progress of Zig, which after 11 years has reached version 0.16 and releases are sometimes characterized by major breaking changes. "When we tag 1.0 it will be a true, uncompromising labor of love," he said. It will also be a backward-compatibility promise, he said, whereas in pre-release the team can continue to improve the language without that constraint. The aim, he said, is to create a language for the next 50 years. Why did Zig move from GitHub? "GitHub simply stopped working for us," he said. "We're here to write software. If our continuous integration server doesn't work, we need to find one that does." Zig migrated to Codeberg, which Kelley said was "essentially a clone of GitHub, so it was an easy transition to make." He also likes that Codeberg is a German nonprofit, on the grounds that "I find nonprofits to be a more stable business than startups or corporations." Zig itself is funded by the Zig Software Foundation, which is a US 501(c)(3) nonprofit. A contentious aspect of Zig is a  decision three years ago to "fully eliminate LLVM, Clang, and LLD libraries from the Zig project," though the Clang compiler will remain. According to Kelley, "you want to avoid a dependency for your core product. We have done this with LLVM, so we're in the process of rectifying this mistake." If JetBrains was hoping for endorsement of its tools, it was disappointed. "I have never used the JetBrains products because it's closed source," said Kelley. He uses a terminal and Vim (Vi improved) for coding, he said. One thing we did not learn is when to expect Zig 1.0. According to the release notes, though, "the 0.17.0 release cycle will be short." ® ]]></content:encoded>
                <enclosure url="https://image.theregister.com/?imageId=5247958&amp;width=800" type="image/jpeg" />
                <media:thumbnail url="https://image.theregister.com/?imageId=5247958&amp;width=800" />
            </item>
    <item>
        <guid isPermaLink="true">https://www.theregister.com/a/5246891</guid>
        <link>https://www.theregister.com/offbeat/2026/05/27/microsoft-excel-champ-proves-he-still-has-the-formula/5246891</link>
        <pubDate>Wed, 27 May 2026 12:29:00 +0200</pubDate>
        <title>Microsoft Excel champ proves he still has the formula</title>
        <description><![CDATA[ Diarmuid Early dominates Amsterdam qualifier as competitive spreadsheeting sets sights on Vegas finals ]]></description>
        <category>offbeat</category>
                <lab:kicker><![CDATA[ Offbeat ]]></lab:kicker>
                <dc:modified>Wed, 27 May 2026 16:44:47 +0000</dc:modified>
                <content:encoded><![CDATA[ Microsoft Excel may be where joy goes to die in rows and columns, but this weekend it was the stuff of trophies, prize money, and esports-style drama. Competitive Excel wrangling has become quite a thing. The latest stop was Amsterdam's H20 Esports Campus, where spreadsheet athletes competed across five formats for a share of a €12,500 prize fund and direct seeding into the 2026 Microsoft Excel World Championship Finals in Las Vegas. Reigning champion Diarmuid Early of Ireland was back at the keyboard. The competition was divided into multiple threads: the main event, a relay (where a three-person team shared one workbook), mixed doubles (where two competitors worked in parallel on linked problems), a student challenge, and a bracket-style "Mega Elimination." Early took home three trophies, winning the Main Event and Mega Elimination, and as part of Team Titanic, which won the Team Relay. Early's success means a direct seed to the Microsoft Excel World Championship semifinals, while Sergio Trifiletti, second in the Main Event, and Alexander Freedman, third, earned quarterfinal spots. The prize for first place in the Main Event was €2,000, being a member of the winning relay team was worth €500, and the Mega Elimination winner was awarded €500. A Microsoft post on the event called Early's achievement "an unprecedented showing on the Excel esports circuit." Excel was not the first commercially available spreadsheet – VisiCalc has that honor, at least on personal computers – but Microsoft's grid is the one that stuck. Having seen off Lotus 1-2-3, the veteran package Early told The Register he had never used, it continues to lurk beneath countless corporate financial models. That said, Microsoft has not always been Excel's best friend. A recent attempt to park Copilot in the middle of the action via a floating button that obscured spreadsheet content went down about as well as a circular reference. The company has since backed off, adding an option to remove the user interface element. The Las Vegas event runs from November 30 to December 2 at the HyperX Esports Arena, and has a $100,000 prize fund. The online qualification round opens on September 26. ® ]]></content:encoded>
                <enclosure url="https://image.theregister.com/?imageId=5246919&amp;width=800" type="image/jpeg" />
                <media:thumbnail url="https://image.theregister.com/?imageId=5246919&amp;width=800" />
            </item>
    <item>
        <guid isPermaLink="true">https://www.theregister.com/a/5246575</guid>
        <link>https://www.theregister.com/software/2026/05/26/california-may-let-linux-bypass-age-check/5246575</link>
        <pubDate>Tue, 26 May 2026 21:49:21 +0200</pubDate>
        <title>California may let Linux bypass age check</title>
        <description><![CDATA[ Exemption in amendment offers relief to open source software makers ]]></description>
        <category>software</category>
                <lab:kicker><![CDATA[ Software ]]></lab:kicker>
                <content:encoded><![CDATA[ The kids are alright. Open source operating systems like Linux and FreeBSD may soon be exempt from California’s app and OS age verification requirements. Last October, California Governor Gavin Newsom signed the Digital Age Assurance Act (AB 1043) into law, which establishes age verification obligations for operating system providers, covered app stores, and application developers. Those distributing operating systems must provide "an accessible interface at account setup" for the user to indicate birth date, age, or both. The act, authored by Assemblymember Buffy Wicks (D-Oakland) and Senator Tom Umberg (D-Santa Ana), aims to protect children from online risks such as cyberbullying, sextortion, and mental health harms. It takes effect January 1, 2027. After AB 1043 was signed, Wicks in February introduced AB 1856 as an amendment to the law. Several changes have been made to the bill since then, the most salient for open source projects being the version published on May 18, 2026. That version includes the following additional language that creates an open source carve-out: (2) "Operating system provider" does not mean a person or entity that distributes an operating system or application under license terms that permit a recipient to copy, redistribute, and modify the software. So if the proposed amendment gets approved, Linux vendors should be off the hook for implementing age checks upon distro installation and launch. Whether that will apply to companies like Valve, which ships its proprietary Steam Client with its Linux-based SteamOS, isn't clear. MidnightBSD in February briefly included a clause in its license that banned California residents from using the operating system. But the following month, project developers set about exploring an age verification mechanism. The Electronic Frontier Foundation has been critical of AB 1043 for outsourcing censorship to app developers. Such rules harm "users' and developers' right to free expression, their digital liberties, privacy, and ability to create and use open platforms," the advocacy group said in March. "It also, perversely, entrenches the dominance of major operating system developers and device makers." At least 25 state age verification laws have already been enacted, and a West Virginia age verification law is scheduled to take effect next month. Colorado lawmakers have approved a state age verification bill that currently awaits approval from the governor. According to Carl Richell, founder and CEO of Linux laptop maker System76, it includes exemptions for open source operating systems, applications, code repos, and containers. In an SSRN paper released earlier this month, George S. Ford, chief economist of the Phoenix Center for Advanced Legal and Economic Public Policy Studies, a free-market think tank, expressed skepticism about the utility and cost of age verification laws. "The effectiveness of these laws at protecting minors is questionable, as motivated teenagers – who already use VPNs to bypass school filters – can easily circumvent age restrictions," he wrote, adding such laws will certainly impinge upon the First Amendment rights of adults by unduly burdening speech. Santa Clara University law professor Eric Goldman on Monday published a blog post looking at the impact age verification has on website traffic – the balk rate or refusal rate. The rate varies but for some sites like Pornhub, it can be as high as 99 percent. Citing the Supreme Court’s assertion in Free Speech Coalition v. Paxton that “adults have no First Amendment right to avoid age verification,” Goldman argued that courts may not treat high balk rates as constitutionally significant, even if credential checks make online movement more constrained and costly. "[W]hoever is doing the centralized authentication won't do it for free," Goldman writes. "A small number of entities are poised to extract monopoly rents by taking a cut of this government mandated process." In 2021, the Age Verification Providers Association estimated that within the next 10-15 years, annual revenues from selling age verification services to OECD countries would reach about $11.4 billion (£9.8 billion). And that was before the US states began implementing age verification laws. ® ]]></content:encoded>
                <enclosure url="https://image.theregister.com/?imageId=254050&amp;width=800" type="image/jpeg" />
                <media:thumbnail url="https://image.theregister.com/?imageId=254050&amp;width=800" />
            </item>
    <item>
        <guid isPermaLink="true">https://www.theregister.com/a/5244877</guid>
        <link>https://www.theregister.com/software/2026/05/25/openbsd-79-arrives-a-diamond-in-the-rough-proud-of-every-sharp-edge/5244877</link>
        <pubDate>Mon, 25 May 2026 09:33:00 +0200</pubDate>
        <title>OpenBSD 7.9 arrives, a diamond in the rough proud of every sharp edge</title>
        <description><![CDATA[ Sixtieth release adds more cores, delayed hibernation, and basic Wi-Fi 6 without losing its ascetic streak ]]></description>
        <category>software</category>
                <lab:kicker><![CDATA[ Software ]]></lab:kicker>
                <dc:modified>Fri, 22 May 2026 14:21:01 +0000</dc:modified>
                <content:encoded><![CDATA[ HANDS ON Even after 60 releases, to borrow Carlsberg's slogan, OpenBSD is probably the most secure FOSS Unix-like OS in the world. OpenBSD 7.9 arrived just a couple of days after project lead Theo de Raadt's birthday. Our congratulations to both. The last four months or so have seen the fastest succession of security issues in Linux that we can remember in the project's existence so far, but OpenBSD sails on serenely. Back in March, Anthropic announced that its Claude Mythos LLM had found a successful OpenBSD attack – but it wasn't a hole. A TCP/IP packet with malformed Selective Acknowledgement options could crash the kernel. This was a real problem, and the bug that caused it went back 27 years, but it doesn't let anyone in. The OpenBSD developers had already included a fix for the bug two weeks earlier, so OpenBSD 7.8 users would get it the next time they ran sysupdate, and it is of course fixed in this version. The new features in version 7.9 are relatively modest. On x86-64 machines – which it terms amd64 – 7.9 now supports a maximum of 255 processor cores, and fixes a bug on machines with over 512 GB of RAM. It can also handle up to 52 partitions per disk. Internally, there can be up to 64, but the limit is now the number of lowercase and uppercase letters of the Roman alphabet, which it uses in labels. On x86-64 and Arm64, the CPU scheduler now understands heterogeneous CPU cores with different performance levels, and can assign processes to four different performance levels described by the letters S-P-E-L, denoting SMT, performance, efficient, and lethargic. This should improve power management, and another feature called "delayed hibernation" can also help. Rather than letting a suspended laptop simply turn off if its battery runs out, when power levels get very low, the machine will wake up then immediately hibernate – a process that ends with it turning completely off. OpenBSD still doesn't have a journaling file system. It uses FFS2, an improved version of the original Berkeley Fast File System developed by Kirk McKusick. This used to include a performance enhancement called soft updates (McKusick's own explanation) but these were removed in 2023. That means that turning off a running machine without shutting it down could cause disk corruption. Delayed hibernation will help prevent one cause of that, at least. The release announcement also lists other changes, including improved support for RISC-V boards, basic support for Wi-Fi 6, the graphics driver stack from Linux kernel 6.18, and even more optimizations to the already-low-latency sound driver stack. There are various tweaks and bug fixes for the various RISC platforms it supports. Version upgrades include LibreSSL 4.3.0, OpenSSH 10.3, and many improvements to the Berkeley Packet Filter (bpf) and Packet Filter firewall (pf), including source and state limiters. Desktop use is not the primary goal of OpenBSD, but you certainly can. It includes multiple window managers and desktops, as documented in its handbook – although this is slightly out of date. Version 7.9 includes GNOME 49, KDE Plasma 6.6, MATE 1.28, Xfce 4.20, LXQt 2.2, and various more minimal window managers. It has its own X11 server, Xenocara, based on X.org 7.7 and Xserver 21.1.21, but you can also run XLibre with some manual effort, and some desktops support Wayland. There is also a downstream project to build a live bootable medium called FuguIta, although it hasn't caught up with the new release just yet. OpenBSD releases are each accompanied by a unique banner painting and theme tune. This time, it's a swinging jazz instrumental called Diamond in the Rough [MP3], which we really enjoyed. It's by Bob Kitella, who along with de Raadt is one of the team at the Alberta internet exchange YYCIX. Calling OpenBSD a diamond in the rough seems quite appropriate. It does have some significant gaps in its functionality, but it is small, clean, and secure. We very much enjoyed a recent essay on ascetic computing by Dave "Ratfactor" Gauer, in which he discusses why his OS of choice is OpenBSD. Out there in the chaos of the open source communities on the social networks that this vulture visits, we often encounter great resistance when we tell people that they're experiencing problems because of their poor choice of equipment. For an easy life and a reliable computing experience, we advise against wireless devices (peripherals or networks), Bluetooth audio devices, and so on. The vicissitudes of Nvidia support on Linux have long been well understood, and eloquently conveyed by Torvalds himself. Avoid this stuff, use devices with plain old cables, and things tend to work more easily and more reliably. Here, we are coming to appreciate the OpenBSD stance on Bluetooth, for instance: it simply does not support it at all. This approach reminds us of the way that Python sliced through the Gordian knot of indentation styles. For instance, this C style guide [PDF] identifies 14 named indentation systems. Python dispenses with all that by making indentation syntactically significant, ending the flame wars at a stroke. Of course, many veterans howl their dismay and rage at this – and yet Python consistently ranks as the world's favorite language, over and over and over again. OpenBSD cuts through some of the complexities of Linux and the other BSDs in a broadly similar way. There has been some controversy recently over OpenBSD's inclusion of code written with AI assistance. The OS includes the tmux terminal multiplexer – and recently, the tmux developers accepted some LLM-assisted code, including the recent DECSET 2026 support. This is now also in OpenBSD, and it's not the only one. No LLM-created code has been committed directly into OpenBSD as yet – and it looks unlikely, if only for copyright reasons, as de Raadt laid out in March. The tmux changes were grandfathered in indirectly because OpenBSD has included tmux in its base system since 2009. We've looked at the changes and they seem small, clean, and innocuous to us. Arguably, the objection is an ideological one of purity. We fear that OpenBSD may end up on the Open Slopware list we mentioned in January. When we reported recently on Fedora and Ubuntu's AI moves, we mentioned the Stop slopware site and the No-AI Software Directory. This probably means OpenBSD won't appear on the latter either, but we suspect that the team will not care. OpenBSD version upgrades are relatively simple, straightforward, and well documented. So, to take 7.9 for a spin, we first tried it in a VirtualBox VM. Although it's a small OS, it wants a large virtual drive because by default it creates nine separate partitions, and because of their different permissions, they're a key part of the OS's enviable security. Worse still, their sizes cannot be dynamically adjusted. Since the installation program is a very low-tech plain-text affair, it offers no help with customizing the layout: if you don't like its proposal, then you must devise your own completely from scratch. It really would help massively if OpenBSD had some kind of simple Logical Volume Manager. Give it enough space, though, and installation goes smoothly. We also tried on the bare metal of an old Lenovo ThinkPad X220, with its own dedicated 128 GB SSD. This threw up an interesting wrinkle: it found the machine's Wi-Fi controller no problem, identifying it as an Intel Centrino Advanced-N 6205 – but because the necessary firmware was not included on the 761 MiB ISO download, it couldn't activate the device, even though it let us enter our WLAN credentials. That's a problem, as the installer defaults to fetching the installation file sets from the internet. We plugged in an Ethernet cable, and then installation continued and finished successfully. The installer automatically installed the required firmware package, so on our first reboot, the Wi-Fi connection came online all on its own. Installing this vulture's preferred desktop environment was as simple as logging in as root and entering pkg_add xfce. Selecting it is not quite so easy, though: OpenBSD's display manager, xenodm, lacks the ability to choose a desktop environment. To fix that, we needed a one-line, two-word script: create an ~/.xsession file containing exec startxfce4, and that was it – a fully working graphical desktop. We added a second monitor, and it was detected, added, and enabled automatically, and we could set it to portrait mode in Xfce's display settings. Although the X11 section of the OpenBSD Handbook says that KDE's recently replaced SDDM is available, as far as we can tell, it has been removed from 7.9 – as has Ubuntu's LightDM. Even so, just saying "yes" when the installation program asks if you want GUI results in a working Fvwm 2.2.5 environment. The Reg FOSS desk has been exploring OpenBSD since version 7.1 in 2022, including 7.2, 7.5, 7.6, 7.7, and 7.8. It's still not an easy OS to install, but if you can dedicate a computer to it, installation is much easier. We recommend avoiding complexities like dual-booting and multiple drives. As a small bonus, it boots and installs perfectly from a Ventoy multi-OS USB key. OpenBSD still supports x86-32, there's no trace of systemd and never will be, and if you really want GNOME or KDE, you can have them. Bringing up a GUI-based system remains substantially easier than it does on FreeBSD. If you're prepared to obtain the hardware it wants, rather than hoping that it will support whatever kit you happen to already have, this is an excellent way to improve your Unix skills – as well as starting to enjoy computing again, free of the distractions of shinier FOSS OSes. ® ]]></content:encoded>
                <enclosure url="https://image.theregister.com/?imageId=5244954&amp;width=800" type="image/jpeg" />
                <media:thumbnail url="https://image.theregister.com/?imageId=5244954&amp;width=800" />
            </item>
    <item>
        <guid isPermaLink="true">https://www.theregister.com/a/5245365</guid>
        <link>https://www.theregister.com/ai-ml/2026/05/25/google-has-seriously-leaned-into-ai-enshittification-lately/5245365</link>
        <pubDate>Mon, 25 May 2026 01:00:00 +0200</pubDate>
        <title>Google has seriously leaned into AI enshittification lately</title>
        <description><![CDATA[ Could the Chocolate Factory's mission to reshape the web backfire? ]]></description>
        <category>ai + ml</category>
                <lab:kicker><![CDATA[ AI + ML ]]></lab:kicker>
                <dc:modified>Mon, 01 Jun 2026 20:54:46 +0000</dc:modified>
                <content:encoded><![CDATA[ KETTLE Google I/O has ostensibly been an AI show for a few years running, but this year's announcements have taken the cake, which Google seems all to happy to let its users eat as it reshapes the web. On this week's episode of The Kettle, host Brandon Vigliarolo is joined by  El Reg senior reporter Tom Claburn and open source reporter Liam Proven to discuss how Google's bevy of AI announcements, and declaration that we're entering the era of AI search, might not play well with customers. From an enlarged AI mode, to AI ads stuffed into AI answers, and pushing AI devs onto closed-source tools after shuttering open-source ones, Google is leaning hard into its version of the future of the internet no matter what users might think, and we wonder whether that might finally crack Google's stranglehold on the web. You can listen to The Kettle here, as well as on Spotify and Apple Music, or read the transcript of the latest episode below. It's been lightly edited for clarity. Brandon (00:04) Welcome back to another episode of The Register's Kettle Podcast. I'm your host Brandon Vigliarolo, and you've likely heard about this week's topic if you've paid any attention to the internet in the past week. Google said at its annual I/O event that it's reinventing search for the AI era. But from an outsider perspective, it seems a lot more like Google's leaning into AI as an excuse to reshape the web and Gemini's image, regardless of how that might affect access to the open web. Unpredictably, there are a lot of people calling foul over that and other recent AI moves made by Google. With me to discuss this is El Reg Senior Reporter Tom Claburn. And joining us for the first time on this iteration of the kettle is our open source guru Liam Proven. Thanks to both of you for being here. Thomas Claburn (00:45) Thanks. Liam Proven (00:46) Thank you. Brandon (00:46) So hey, Google's AI-ification of search was the big news to come out of I/O this week. Tom, you tuned into the keynote and wrote about this. So what exactly did Google say it's going to do and why is everyone so up in arms over this? Thomas Claburn (01:01) I mean, it's just more encroachment of AI into search and they, you know, they have their AI Overviews, which are the little summaries that they put up on top of search results. And then they also have separate thing that's very similarly named, but different called AI Mode, which is available through a tab and you click on it that's sort of a deeper version of AI, I think it ties into some, Google knowledge graph and it has sort of a broader thing, but you often get similar results, but basically they're going to be showing more of the AI Overviews and, it's not always clear when these happen, but basically for longer queries, it's more likely to be handed off to an AI model. Brandon (01:44) Mm-hmm. Thomas Claburn (01:45) And it's a problem for a lot of people because people's relationship with Google began with: you go to Google, you find stuff, and then you leave. And increasingly, it's you go to Google and you're stuck there like it's a tar pit. And you're just trying to figure out where did they get this information? And they'll put up a summary. And of course, they have the disclaimer, well, you know, maybe it's not accurate. You'll have to check on that. How are you going to check on it? I'll go to the links that we didn't show you. It's, you know, people I think are a little bit – I mean, part of it is just people don't like change, but part of it is just AI really is not the right answer for a lot of things, at least in my opinion. I think there are certain kinds of queries that it can be useful for. And I think that largely though, if people are going to look for documents, they need to be able to find reputable sites and be able to make trust decisions. And a lot of that information is getting obscured or put into little teeny citation chips that you have to click on to figure out, where is this information coming from? Brandon (02:49) Yeah, and sometimes when you click on one site, it'll give you four or five links and be like, well, here's the sources we use to compile this information. Like a lot of times, I'll admit, I do use the AI Overviews every once in a while when they pop up, especially for simple questions like on my smartphone or something. But they'll give you, cite their sources and you click on them. But sometimes that's just as big a pain in the butt as assuming that the AI Overview is just correct. I'd much rather just have a list of blue links, which Google did clarify to me and to Avram, our US editor, earlier this week, that traditional search engine result pages are not going away. Thomas Claburn (03:24) Yeah, they're not going away. They're just going to get buried under more AI. You have to work harder to find them. And then there was some other interesting stuff too, where their Gemini Spark, which is their agent... in the Gemini app, they're also going to be pushing these long-running AI tasks that you'll be able to do, and they're eventually going hook it up to the regular Google account or search or whatever. So you can basically run a chron job with, you know, an AI model essentially, to go do things for you. And I think that the think they talk about it for is shopping. It's gonna, of course, plan your travel itinerary and do stuff for you in the background, and somehow you're gonna be happy with results. It's not clear how you're gonna pay for that because someone has to run this stuff, maybe this all comes out of the hide of advertisers who are gonna sort of get shoveled into these results, who knows? Brandon (04:20) Right, and that's actually kind of segues really well into one of the stories that I wrote about I/O this week, and that was Google's new AI advertisements that they're kind of injecting in. They... I see we're doing audio, but I see Liam on the other side of the camera just putting his hand in his hands, you know, my God. We talk about where the cash is going to come from from this, and it's obviously going to come from this, right? There were two particular kinds of AI advertisements that Google said they were going to add soon to AI mode. There were some ads that were going to be basically in line. If you ask a query, you get your responses back from the AI. In that list of results is going to be ads. And Google said they're going to be at the bottom of the list, but they're still going to be presented in line. And I think there might be some indicator about them being a sponsored element of the post, but they're still putting them in line with results that are ostensibly grabbed from the web and are supposed to be factual. The other one that I found a little more concerning personally was conversational discovery ads. So basically the way Google described this is, you ask a complicated question and it will use Gemini to figure out what products you need to solve your problems. And the example they gave was, oh well, your house smells kind of musty and you want to make it smell more like a spa. Well, I feel like if I were to go onto Google right now and type in, my house smells musty and I don't want it to. Some of the first things you might get are things like, put some baking soda in some water, make a 50-50 mixture of vinegar and water, and you can deodorize and clean for pennies on the dollar. But Google sees this as a way to inject products in front of you.... I was picturing going in there and asking for tips on cleaning my house and deodorizing it and getting a whole bunch of ads for $20 reed diffusers, expensive plug-in units. Liam Proven (05:55) Mm-hmm. Brandon (06:08) That's how I see this, right? And I don't know if that's entirely correct, but Google's not doing a lot to kind of say that that's not the case. Liam Proven (06:17) I feel like the great prophet Cory Doctorow kind of nailed this a couple of years ago now with this word, enshittification. I was baffled when Google announced that it was going to start deliberately degrading search results in the interest of keeping people on the page and on the site longer. And it feels like they're not leaning in, they dived off the board and jumping in, pinch the nose and throw yourself in. I do not see how this is going to pay long term, but, on the one hand maybe there's some genius there with playing playing four-dimensional chess, maybe they're just... they've drunk the Kool-Aid and cannot imagine anything else now Brandon (07:05) Yeah, it just feels like an attempt. I mean, it was the same thing with Google saying that they were introducing, I think, some commerce protocols earlier this year that were designed to basically allow Gemini to check out for you. So you don't even need to go to a company's website to buy a product now. You can do it all right through Gemini. So that, again, that's starving a company of web views so that Google can make a few more cents on a transaction. Thomas Claburn (07:28) Yeah. Brandon (07:29) And I don't see how this is any different, right? It's injecting more ads, getting more things in front of you, and burying actual web results below this in the hopes that you never get to that point. Thomas Claburn (07:36) And the one thing it's going to incentivize is that everyone who actually wants to use an AI model is going to think, how can I use this to block ads? How can I use it to get this stuff out of my Google search results and get something that's closer to whatever, some kind of neutral or less commercial standard? At least if you put up with it, if you don't just turn it off entirely and think, I'm going to figure out – I'm going to go back to Yahoo and get a list of curated sites and just stay there. Brandon (08:06) I think about it sometimes in terms of, you know, well, Google Search was never giving me an objective view of what's on the internet, right? It's always filtered through Google's algorithm or whoever's. But I feel like there's a difference between filtering it through an algorithm and making me do the legwork and just assuming that whatever Gemini is serving to me is going to be exactly what I need because Google thinks that's what I need. Liam Proven (08:31) Maybe you take a, what's the expression, a 30,000 foot view, but I think we've got to go a lot higher and take a low earth orbit view. This is going to be very good news for the wider software community as it drives improvements in ad blocking technology, Google-free browsers, Google-free search engines, Google-free anything, please, anything that can get this stuff out of our face. Thomas Claburn (08:56) Europe is already partway there with the sort digital sovereignty stuff. I mean, this is just another sort of data point in the rationale for moving. perhaps we'll finally see some innovation where Google kills its own search business. And it's not like search was doing so well anyway. Even before the AI boom, was a lot of complaints that there were just lacks about policing, spam farms and things like that. There was a lot of lifting you had to do even just as a 10 blue links user to sort through the junk. And if they really cared about delivering quality editorial to people, the web would look a lot different. Brandon (09:35) You just wrote an op-ed, Tom, that kind of covers some of that, right? You asked Google's own AIs why Google search results were getting worse, and it pretty much was like, yeah, hey, they are. Mea culpa here, you know, it was... Thomas Claburn (09:39) Right. It's unfair, but it's also, kind of telling that this is what we've come to where, we're going to source stuff off of a couple of Reddit opinions and blogs. And then, when you ask it, when you frame it in a nice way, "why is Google search great?" It goes to Google's own blog posts to source that. I guess that works for some people, but it's really just a poisonous media ecosystem. I who wants to even be a part of that? I mean, I think all of this just drives a lot of sane people away. And the only people who are left are gonna be sort of hucksters and grifters and people who are trying to game the system for whatever commercial intent they have. Liam Proven (10:27) It is. Yeah, it's going to. think one possible effect will be driving the creation, the fragmentation of the Internet, maybe not exactly layering, which is what I thought might be happening a decade or so ago. But in the same way that there are sites like Conservapedia and so on that try to present a U.S. right wing Moral Majority type view of Wikipedia, that there will be a fragmentation into the AI-driven web and the little indie, we're trying to keep this out, AI-free web. There already was this attempt the other Gemini, Gemini the protocol, it was launched about a decade or so ago now. Brandon (11:19) Yeah, really quick, remind us what that is? Liam Proven (11:22) It's an extremely lightweight protocol and markup language for serving pages of hyperlinked content, that's not the web. And you needed to run a Gemini server and you needed to use a Gemini browser to access it. And it gave you kind of like a markdown view of the web. So you got, bold, italic, and underline and nothing else ... you have no color, you have no fonts, you have no images and really, really stripping back the online hypertext experience to like an early nineties view. It's kind of faltered in recent years, but it's people choosing this very minimalist and stripped-back experience and much as it seems, I think, to baffle marketing executives, I think a lot of people would choose an un-augmented AI-free view of the web these days. Brandon (12:18) I mean, just think about, good Lord, the videos coming out of college graduations recently, right? know, everyone, I think Liam, you mentioned earlier, everyone except Woz has been booed to hell and back in the past couple of weeks for making AI claims. I think there's a lot of dissatisfaction with it, right? I think you're right. I think there'll be a lot of people. Thomas Claburn (12:36) The irony is that this should be the moment to shine for the social media networks that are notionally about people and connections, which haven't been with Meta and Facebook. I mean, it's turning into the same AI slop engine as everything else and is driving people away. There's theoretically room for some kind of network that humans can interact with because it's just not pleasant to be with bots. I mean, nobody wants to get AI communication. No one wants to deal with it. And you just cannot operate in the same space as a bot. They operate on a velocity that people can't deal with. And it's just frustrating. So I don't know how people are going to want to participate in this. Liam Proven (13:14) I've gone through a little unexpected voyage in recent months because oddly enough, as a professional writer, and I also read a very great deal and have since childhood, I kept reading people saying, "this text is AI generated. I'm not going on any further because I read the first couple of lines and I just knew. It's full of AI tells." And so I started asking people, you know, what are the tells? What are the signs? Oh it's, it's, it's just obvious. It's redolent. Okay. But can you give me a clue here? You know, what, what are the things that give it away for you? And I have not yet been able to get anybody when pressed to give me a nice clear list. You know, a lot of it boils down to em-dashes and I've been using them for years, but Brandon (14:04) I wasgonna say I love the dash, so it really kind of throws a wrench in my work. Proven (14:06) Right? You know, I learned there are a handful of actual fairly concrete things, you know, the "not only, but also" kind of structure that the bots do overuse. But I'm seeing people going, "I'm not going to read that because it's obviously bot-generated." And then somebody else pops up and goes, "Hi, I wrote that. That's my site. I don't use any bots. That was all just me." One guy recently I was reading said, "Look, I've got the git history with the 19 commits, because it's not a very long site, you know, as I wrote it. Would you like me to show you the process by which I wrote every word?" And of course the guy challenged goes, "Well, I don't know. I'm not sure I believe you." And he's like, "Okay. So you've gone from a certainty to doubt. I guess I'll take that." But I watched with interest the relaunch of the Digg social network over the last year or so. And Digg was very much like Reddit about 15 years ago, coming up to 20 years ago, it was a site with lots of special interest groups where you posted interesting stuff, but was driven by threaded conversation. Brandon (15:12) Yeah, I remember Digg. It was great. When it went under I just moved to Reddit. Liam Proven (15:20) And, and, the guy, I think it's Kevin Rose that owned it, got it back and relaunched it very much in the old model, but without the ability to create your own groups and stuff. So he was trying to keep it a bit smaller. And I joined, because I had a Digg account in the day and thought, this is strangely bland and anodyne. You know, it's full of people being nice to each other and saying nice things. And it seems to be largely content-free. Well, it shut down again earlier this year. And he said, "We got infiltrated by bots. We got loads and loads of people posting bot-generated content until a lot of the discussions were bots talking to other bots." That would kind of explain what I saw, you know, but I'm still on Twitter, call it Twitter. I'm still on Twitter and I have a block list, which is like six pages long now of words and phrases. And it makes it kind of tolerable, but it is very odd to watch the interchange and they've changed it recently so that you can block accounts, block and mute accounts which are serving ads and yeah, if those are paying, paying advertisers, they ignore your blocks. You get them anyway. ⁓ all right. Yeah. Brandon (16:30) Yeah, of course. I have the same thing on Reddit, right? I've blocked multiple advertisers and I go in to see, well, why am I seeing this? I want to block this account again. It's like, no, you already got this account blocked. And it's like, well, then why am I seeing your ad? Liam Proven (16:39) Yeah, yeah. And yet, you know, it's still actually quite lively and there's a lot of discussion and there are still interesting people and some of the interesting people I followed years ago still are still posting and the discussions are still good. Some people are choosing this experience and it's not just because they're paying for it. Some people will choose this experience for reasons that escape me. Thomas Claburn (17:01) People build up, you know, it's the follower structure. You build up an audience and it's costly to rebuild that. So a lot of people have stuck with X and there are now political reasons to stay with X and a lot of people have done that. And even journalists – I mean, I would have loved to given up my X account, but it's useful because there are still people who post interesting things. Brandon (17:28) Yeah, I don't really post on there anymore, but I still have my account. Thomas Claburn (17:30) Yeah, it's worth it for sort of source finding. People, I think that they flee the AI influence when they see it. If they don't leave the site, they figure out a way to filter it. So I think it's going to be a very difficult few years as some kind of new equilibrium emerges because the old sort of systems where people and bots mix is just not satisfying to people. Liam Proven (17:54) And as X declines, I'm watching Mastodon, the Fediverse grow and get more interesting and get more feedback and interesting discussions. I still find Lemmy kind of a pain to deal with. I don't really like the presentation, which is like the activity pub-driven version of Reddit, kind of, but I'm getting lots of interesting comments and feedback and, I don't like this word, but engagement. People are engaging on Mastodon in a way they used to on Twitter and some of the other sites. Brandon (18:26) Fleeing before AI, and hopefully it won't come to those platforms either. But back to some of the things that Google's been doing, and I guess we can speak more to the alienation aspect here, Another story that I covered this week out of I/O was that Google – a lot of people probably are familiar with the Gemini command line interface, allows you to at least use Gemini to look at code and do some various programming tasks. And that's an open source tool that Google's had for, I think, about a year now. But in classic Google fashion, they're deprecating it in favor of the new one that they announced at I/O this week, Antigravity CLI, which has, you know, some, I think, feature improvements over Gemini CLI, right? Gemini CLI is Gemini. Antigravity is, I think, a little more model-agnostic. But it's closed source. And that means that basically anyone who was using Gemini CLI is going to not be able to use it come June 18th. It's just going to stop working. Google's not ending their maintenance of Gemini CLI. They're just restricting it to high-tier enterprise customers. So it's still there. It's still an open-source product. You've just got to pay to use it now. I found it interesting that a lot of developers, when I was reading some threads about this, were saying that they were particularly upset about the fact that they felt like they had spent their time and their effort to help improve Gemini CLI through bug reports and things on GitHub, right? And now all that work is essentially being closed-sourced in a new product and sold back to them. You know, I wonder again, right, is this Google kind of leaning into another AI product that's just gonna piss more people off? Thomas Claburn (20:00) Right. Well, I mean, the first lesson is never, never bet on a Google product because they kill them off mercilessly. It's also, the amount of damage that the AI has done to the open source community, we're going to be dealing with it for years...I've had a couple of projects where I think like, do I even want to make this public because I'm just going to get, if anyone uses it, I'll just get AI bug reports. And in a sense, all of open source has been captured in these models anyway. And you can just ask the model to regenerate all this unlawfully captured labor that is latent in these things. And we haven't figured out a way to deal with that. I hope that some of the software lawsuits make progress because it's really transparent that people's labor has been captured in ways licenses did not condone or anticipate. People are just reselling that labor at increasingly high prices. Brandon (21:00) Yeah. Liam, you cover open source stuff a lot. What has the developer community been saying about AI and its influence. Liam Proven (21:06) Again, I think. Obviously this, has to be a super generalization, but it, seems to me that it's, it's splitting and factionating. And on one side, there's a group of people who embrace the tools, say that it's delivering unprecedented levels of productivity and so on. And on the other side, on the, the other faction, there's a group of people saying, no, we will not allow this anywhere near any product that we use, run, develop. There are a handful of people who are kind of still in the middle like looking from side to side and I recently wrote about the new version of OpenBSD and it's faced such a problem which is that OpenBSD incorporates tmux, a text mode terminal multiplexer. So you can have windows in your terminal and different stuff going on. And tmux started allowing Claude-assisted code contributions. And that means they got grandfathered into OpenBSD. But the OpenBSD project has said we won't allow AI-generated code because we can't copyright it. We can't put a firm license on it because we can't say where it came from. So on the one hand, we can't allow you to contribute authored code. On the other hand, here's an externally maintained project, which is using AI-authored code. So they're kind of stuck in the middle. The most interesting study I've seen on this where somebody tried to put numbers on it was from an organization called [METR ]. And they published the results of a study they did. They did a controlled trial with a whole bunch of developers given various programming tasks. And one half of the subjects were allowed to use AI tools and the other one not. And at the end of it, they asked the developers, how was it for you? You know, was it helpful? And all the people using bots went, it's great. We estimate it's taken about 20% of the time off the process of developing this feature and getting it working. And then they compared it with the other developers who weren't using any kind of code generation. And in fact, the people not using the tools were 20% quicker. In other words, it feels like you're going faster, but actually you're going slower, but you sit there and watch the code unfold on your screen. I don't know. I've never used any of these tools. I'm an AI vegan. I avoid the whole thing, but you sit there and watch the bot write code and go, well, that was quick. All I've got to do is make sure it works. And four hours later it works, but you could have written it in three hours. And the odd thing is I saw this report and I linked it and shared it and cited it. And then I bothered to go and look at who [METR] is. And [METR] is a pro AI advocacy group. They are AI boosters, but they decided let's get some proof of how much quicker it makes you. Well, we said we're going to do it, so we're going to publish. Respect to them for that. I do wonder if all of the AI assisted projects, the ones that are really leaning in, are going to come up with a nasty surprise. Either this didn't really help and we got a load of stuff we can't debug, or actually this is taking longer, or the price of this tool I'm using just went up, it just gained a zero and it's going to go up again and damn it, I can't afford to use this anymore. Thomas Claburn (25:00) Well, there was just a report recently that Microsoft canceled its internal cloud licenses because of the price hikes. I the price is going to go up and ⁓ the problems aren't necessarily going to be evident right away. Brandon (25:15) I've heard multiple instances of companies basically saying, well, these AI products are doing fine, but at the end of the day, they're more expensive than a new developer fresh out of college. So why would we use them? So it's interesting. So before we wrap up, I wanna touch really quick on something that you wrote too, Liam, it was a couple of weeks ago, but I think it still plays into this whole idea of Google's. AI-ification thing. that Chrome was quietly installing large language models on users' machines without expressive consent. So I understand you can flag that and turn it off. It's an opt-out thing. But is that still happening? Did Google change tack on that after this was reported on at all? Liam Proven (25:55) As far as I know, no, I have not checked because I didn't have it anyway. Every time any app offers me any kind of AI integration, I just turn it off as soon as I can. And I'm not even sure what boxes I ticked in Chrome and when, but when I found out about this, I went looking and no, not on my –I have my Google Chrome profile synced onto Windows and Linux and Macs, and it wasn't there on any of them. But on the one hand, I'm noticing in the open source world, some projects are belatedly embracing this, but they're talking very much about local first, open source models. Let's keep it on your machine, private. There's no risk of any leaks. And actually that is kind of what Gemini Nano was supposed to do. It's a tiny model, four gigabytes, a tiny model that wouldn't fit into memory on an x8632 box, but hey, a nano model that ran on your machine so there was no risk that anything could leak. Well, that is actually a good thing, I guess. Brandon (27:06) Sure, yeah. Liam Proven (27:10) But it did it on phones as well. And now, okay, you know, I'm cheap. I use very low-end, mostly Chinese phones. But you know what? I live on a little island in the middle of the Irish Sea. I fly a lot more than I used to these days. And a couple of months ago, I was about to make a trip and I thought, I'll put some new music on my phone. The only time I use my fancy noise-canceling headphones is on planes. Oh, my phone's full. And this current one doesn't have a card. I have to sync over a cable. So that's weird. I haven't got that much music on it. I discovered about a dozen or 20 feature films on my phone I have no recollection of ever downloading. I've never seen these films. I guess some search term somewhere synced something. I deleted them. Brandon (27:58) Better an MP4 than an AI model, I suppose, right? Liam Proven (28:00) Yeah, but at least it was something I could sit and watch on the plane. I deleted them all, fitted a bunch more albums on there and all was good. But yeah, even with my cheap-ass 300 buck phone, okay, four gig here or there is kind of nothing. But even so, I'd rather that space was for my stuff. If you're going to take that much ask. Brandon (28:23) Yeah, local LLMs are good idea, but ask me first. Thomas Claburn (28:27) It also raises the issue of what's the difference between that and a crypto miner if someone else is using it, external server is using it. I'm fine with providing storage for something that's going to benefit me, but when some entity that I have a relationship with, or don't, is running stuff locally using my storage capacity and my processor for their benefit, I don't know about that. Liam Proven (28:53) Yeah, exactly. It's like one of those proof-of-stake cryptocurrencies rather than a proof-of-work one, you know. OK, I said earlier I don't use any AI tooling. There's one exception for that, which is I do use language translation tools, and I use them quite a lot and I've got a choice of them. I spent nine years living in the Czech Republic, a country with a brutally hard language I still can't read worth a damn. So if they said, like the deal with Firefox, we're going to put a model in your browser, but it'll translate stuff on your device and it won't go to the cloud. Well, okay, that could actually be useful. It's not quite at the point where I could use my phone to translate a menu while I'm offline on a plane or something, but you can see that is not far off. But you know what? I want to know why, I want to know what your what you're extracting from me and what I get in return, and make my own choice. And increasingly that is a choice we're just not getting. I personally do not think that open source products like Ubuntu, like Fedora, including even optionally, open source models, which are privacy first and local and ... No for me, that's not good enough. I don't want that. And I'm not really interested in any product that includes that. And I appreciate they're trying to do the right thing, but I think they are going to be shocked by the level of hostility. Fedora is already backing down from its moves to attempt to become the best Linux for AI development because to their great surprise, there was a user outcry. I think Ubuntu cares rather less about what its community thinks. They just try and do what they think is best. But I think they're going to be surprised by the pushback, as probably Google was. Brandon (30:53) Google obviously, they might be surprised by the pushback, but the question is, will they care? It seems like by and large, Google's – I think it was 2018 when they formally abandoned the "Don't Be Evil" slogan, I think, right? And it just feels like with these announcements lately that they're really just making sure that it's fully whited out and erased from memory here. I mean, I don't know, putting an AI wall between people and the open internet, secretly uploading LLMs to people's machines, forcing people onto closed source products. I mean, any thoughts on what's gonna happen here? Any thoughts, guys, before we wrap this up? Like, is Google gonna face blowback? Are they too big to fail? Thomas Claburn (31:37) Not for nothing is there all this excitement about someone finally being able to sort of take some of Google's business. I the whole internet AI thing took off when everyone saw a weakness in search and said, hey, we can provide something that will break Google's stranglehold. Frankly, this is a story about the years of failed know, antitrust work that, you know, should have been dealt with many years ago and wasn't and so Google basically just controls a large sector of the internet along with, you know, along with Meta now for advertising. And that's starting to break up a little bit, but it remains to be seen whether AI is going to be an advertising medium that's equivalently lucrative to search, but who knows? going to give it a shot, but users are going to get sacrificed in the process. Liam Proven (32:27) It's like an eternal verity of life in the technology market that, you know, if it's true, there's probably an XKCD about it. And there was an XKCD years ago, 1118. "Remember when we prosecuted Microsoft for bundling a browser with an OS? Imagine the future we'd live in if we'd been willing to let one tech company amass that much power." "Thank God we nipped that in the bud." Nobody is too big to fail. And the bigger they come, the harder they fall. I think some mighty industries worth hundreds of billions are going to come to grief over this stuff. And I haven't got lot of sympathy. Brandon (33:08) I guess we'll see. We'll see if this will be enough to, you know, clean some of that enshittification off the walls of the internet in the coming years. Liam Proven (33:16) What a beautiful phrase, well said. Brandon (33:08) But hey, no matter, yeah, no matter if it happens or not, we'll probably still be here and we'll probably still be talking about it the Kettle, so be sure to tune in. Thanks for joining us. ® ]]></content:encoded>
                <enclosure url="https://image.theregister.com/?imageId=5243010&amp;width=800" type="image/jpeg" />
                <media:thumbnail url="https://image.theregister.com/?imageId=5243010&amp;width=800" />
            </item>
    <item>
        <guid isPermaLink="true">https://www.theregister.com/a/5242367</guid>
        <link>https://www.theregister.com/software/2026/05/22/marketing-demanded-it-add-website-feature-that-was-already-working/5242367</link>
        <pubDate>Fri, 22 May 2026 08:31:00 +0200</pubDate>
        <title>Marketing demanded IT add website feature that was already working</title>
        <description><![CDATA[ Techie regrets not taking credit for getting it done with amazing speed ]]></description>
        <category>software</category>
                <lab:kicker><![CDATA[ Software ]]></lab:kicker>
                <dc:modified>Fri, 22 May 2026 15:07:42 +0000</dc:modified>
                <content:encoded><![CDATA[ ON CALL Welcome to another edition of On Call, The Register's weekly reader-contributed column in which you share your stories of absurd tech support situations. This week, meet a reader we'll Regomize as "Hamish," who told us that a couple of years ago he worked at a British retailer where the company's website manager – a member of the marketing team – came up with a brilliant idea that was bound to boost sales: adding Apple Pay to the company's website. Management approved the idea, which duly landed on Hamish's desk – and confused him enormously because the website already offered Apple Pay. Hamish had two pieces of evidence to prove this fact. One was that when he visited the website, he could see an option to pay with Apple Pay. The other was that he worked for the company during the initial push to enable Apple Pay and remembered the project well. He was pretty sure several of his colleagues would remember it too because they worked in management or marketing at the time and did some of the work! Hamish nonetheless went along with the request by chatting with colleagues in IT and the company's finance team, who confirmed that Apple Pay was indeed up and running, and even sending money into the company's coffers. That ruled out the possibility that the site was buggy in some way Hamish had missed, and meant the next step was to ask the website manager why she didn't think Apple Pay was already available. Hamish said the marketer told him she couldn't see Apple Pay as an option when she visited the site. To prove it, she whipped out her Android phone. "It turns out that everyone who thought this was a brilliant new idea and who had bothered to look at the website had done so without using an Apple device," Hamish told On Call. The company's site was therefore not only Apple Pay-enabled, but also capable of detecting users' devices and dynamically presenting relevant payment options. Hamish isn't sure he handled this situation correctly. "Maybe the IT team should have waited a week, said the work was done, and scored bonus points for a speedy delivery," he mused to On Call. "Instead we used the opportunity to show how unaware senior people were of their own pet projects." Have you been asked to fix something that works, or implement something that's already in place? If so, here's something else that already works – clicking this link to send your tale to On Call so we can consider running it on a future Friday. ® ]]></content:encoded>
                <enclosure url="https://image.theregister.com/?imageId=4094195&amp;width=800" type="image/jpeg" />
                <media:thumbnail url="https://image.theregister.com/?imageId=4094195&amp;width=800" />
            </item>
    <item>
        <guid isPermaLink="true">https://www.theregister.com/a/5244371</guid>
        <link>https://www.theregister.com/ai-ml/2026/05/21/deus-ex-machina-half-of-us-christians-trust-ais-spiritual-advice/5244371</link>
        <pubDate>Thu, 21 May 2026 18:36:47 +0200</pubDate>
        <title>Deus ex machina: Half of US Christians trust AI's spiritual advice</title>
        <description><![CDATA[ AI sycophancy + spirituality = uh oh ]]></description>
        <category>ai + ml</category>
                <lab:kicker><![CDATA[ AI + ML ]]></lab:kicker>
                <dc:modified>Wed, 03 Jun 2026 11:26:45 +0000</dc:modified>
                <content:encoded><![CDATA[ Who needs a minister when you have an LLM? America’s Christian population appears to have found God in precisely the place you’d expect a manifestation of the divine to be spotted in 2026: Amid AI chatbot responses. A survey of Americans published this week by Evangelical polling outfit Barna sought to discover what Christians thought about AI’s ability to serve as a spiritual mentor, and the split is surprisingly even: A full 48 percent of practicing US Christians told the organization that they trusted AI’s advice to aid their spiritual growth. Potentially more surprising than that, 34 percent said spiritual advice dispensed by an AI was just as trustworthy as what they'd get out of a flesh-and-blood pastor. That share rises, unsurprisingly, among younger Christians, with 39 percent of Gen Z respondents and 44 percent of Millennials agreeing that preachers and AI are at trust parity. Pastors themselves, it likely won’t surprise you to learn, are splitting sharply from their flocks on the matter of AI’s ability to fill their roles in the lives of congregates, with just 12 percent saying they agree that AI can help people grow spiritually. That said, there’s a pretty serious tension among American Christians when it comes to AI. At the same time half say it’s aiding their spiritual journeys, most also expressed concerns about negative effects of AI on spirituality. A full 83 percent of practicing US Christians believe AI is likely to misinterpret scripture, 73 percent are worried AI will cause loss of religious faith, and 72 percent believe that AI is beginning to act as a replacement for God and earthly spiritual leaders. “Christians say they trust AI with spiritual growth, and a meaningful share say its spiritual guidance is as trustworthy as a pastor’s—yet large majorities are simultaneously concerned about AI misinterpreting scripture, replacing God, or undermining the role of spiritual leaders, Barna VP of research Daniel Copeland said of the findings, which he called “confounding.” “That level of openness is higher than we might have expected,” Copeland added in the Barna’s report. Worshipping at the altar of Altman AI and religion have been butting heads for the past couple of years, with the Catholic Church particularly outspoken about the technology. The late Pope Francis called on world governments to establish global AI regulations in 2023, as well as calling on people to avoid turning to AI models for moral and ethical decisions. Vatican AI authority Friar Paolo Benanti later accused Silicon Valley elites of playing God with their creations, AI included, noting that “the focus will always be on using AI for profit,” which - according to the good book itself - isn’t compatible with Christianity. That hasn’t stopped some of God’s faithful from creating an AI Jesus and Christian AI platforms, and the new Pope, Leo XIV, has continued his crusade against the technology. "By simulating human voices and faces, wisdom and knowledge, consciousness and responsibility, empathy and friendship… artificial intelligence [could] not only interfere with information ecosystems, but also encroach upon the deepest level of communication, that of human relationships,” Leo said earlier this year. Leo further expressed worry that AI was turning people into “passive consumers of unthought thoughts,” and that’s not even to touch on the fact that AI has a tendency to make stuff up to appease its questioners, potentially leading the spiritually curious into full-blown episodes of psychosis aided by a digital yes-man in the guise of an authority. ® ]]></content:encoded>
                <enclosure url="https://image.theregister.com/?imageId=5244417&amp;width=800" type="image/jpeg" />
                <media:thumbnail url="https://image.theregister.com/?imageId=5244417&amp;width=800" />
            </item>
    <item>
        <guid isPermaLink="true">https://www.theregister.com/a/5243629</guid>
        <link>https://www.theregister.com/oses/2026/05/20/microsoft-rebases-azure-linux-on-fedora-as-fedora-drops-deepin/5243629</link>
        <pubDate>Wed, 20 May 2026 19:24:12 +0200</pubDate>
        <title>Fedora: Microsoft is all aboard, but Deepin is dumped</title>
        <description><![CDATA[ Red Hat’s free distro loses a desktop, but makes an important new friend ]]></description>
        <category>oses</category>
                <lab:kicker><![CDATA[ OSes ]]></lab:kicker>
                <dc:modified>Wed, 20 May 2026 18:21:54 +0000</dc:modified>
                <content:encoded><![CDATA[ Microsoft has announced a new, Fedora-based Linux distro for Azure VMs, while Fedora has consigned the Deepin desktop to the bin. Fedora decided to remove a component maintained outside Red Hat. In the same week, another external company - granted, a slightly better-known one - decided to rebase one of its projects onto Fedora as its upstream distro. It’s the circle of life, or something.    Fedora 💔Deepin Seven years after it added the Deepin Desktop Environment in Fedora 30, Tuesday's FESCo meeting decided to drop Deepin from the distro. The minutes say: AGREED: Retire all packages maintained by the deepinde-sig group The decision comes one year after the project called for a security review of the Deepin Desktop Environment, after openSUSE dropped the desktop following a negative security assessment. We reported on that decision at the time. SUSE asked Deepin for feedback, but didn’t get good enough answers – for which, some months later, the Chinese project issued an apology. Linux Deepin is very much still around: we most recently looked at version 25 in January, and, back in 2023, the project claimed it had passed three million installs of its paid Tongxin UOS desktop edition. It’s a very pretty Windows-like desktop environment, but it never made it to having its own Fedora spin – and it certainly won’t now. Microsoft ❤️ Fedora But as one door closes, another opens. Fedora is still winning new friends and allies, and mere days earlier, there was a surprise announcement at the Open Source Summit North America, which as we write is winding down. On Monday, Microsoft announced a new version of its in-house Linux distro, Azure Linux 4, along with a companion distro called Azure Container Linux. There have been products called Azure Linux for quite a while. It’s based on the much more minimal CBL-Mariner distro, which we tried in 2022. The Register reported on Azure Linux becoming generally available in 2023, and then on the release of version 3 in 2024. We also knew back then that the company was working on turning it into a more general-purpose server OS: we reported on it migrating LinkedIn to Azure Linux in place of CentOS Linux that same year. There isn’t very much information about Azure Linux 4 yet; the broader rollout will be at the Microsoft Build conference next month. For now, all you can do is fill in a form to register your interest. However, the announcement reveals that version 4 switches to Fedora as its upstream distro. It was already based on the RPM packaging tools, hinting at some Red Hat or SUSE heritage in there somewhere. There’s slightly more information about Azure Container Linux. This is a separate distro, an immutable host OS for running containers. The announcement says “Azure Container Linux is based on the Flatcar project.” Flatcar is the continuation of CoreOS Container Linux, which The Reg has covered since it released its first version in 2014. As Linux Weekly News reported that year, CoreOS was based on Google’s ChromeOS, but redesigned to host containers. Red Hat acquired CoreOS in 2018, and then two years later, discontinued Container Linux. It replaced the Google and Gentoo-based distro with a new one based on Red Hat’s own immutable tool chain, called Fedora CoreOS. German FOSS consultancy Kinvolk forked the CoreOS code and continued development under the name of Flatcar Container Linux. Kinvolk was acquired by Microsoft in 2021, but continued to work on Flatcar. Now it seems that, with the announcement of Azure Linux 4 as well as Azure Container Linux, Microsoft has two separate in-house distros: one based on Fedora, and one based on ChromeOS. For now, Flatcar is still trundling along the tracks just fine, but we suspect some future consolidation may be coming down the line. ® ]]></content:encoded>
                <enclosure url="https://image.theregister.com/?imageId=5243659&amp;width=800" type="image/jpeg" />
                <media:thumbnail url="https://image.theregister.com/?imageId=5243659&amp;width=800" />
            </item>
    <item>
        <guid isPermaLink="true">https://www.theregister.com/a/5243605</guid>
        <link>https://www.theregister.com/ai-ml/2026/05/20/bye-bye-gemini-cli-google-nudges-devs-toward-antigravity/5243605</link>
        <pubDate>Wed, 20 May 2026 18:44:43 +0200</pubDate>
        <title>Bye-bye, Gemini CLI; Google's gone and swapped you for a closed-source AI</title>
        <description><![CDATA[ Most users lose access June 18 - unless you’ve got enterprise creds or paid API keys ]]></description>
        <category>ai + ml</category>
                <lab:kicker><![CDATA[ AI + ML ]]></lab:kicker>
                <dc:modified>Wed, 20 May 2026 20:42:11 +0000</dc:modified>
                <content:encoded><![CDATA[ Pour one out for the Gemini Command Line Interface. Come June 18, the open source development agent will stop serving most users in favor of the new Antigravity CLI, and developers aren’t happy that the replacement is far less open than the old tool. Google announced the Antigravity CLI at Google I/O this week, billing it as a way for the Chocolate Factory to unify its efforts in developing a command line interface for AI agents. One of the key arguments Google makes in a post about transitioning from Gemini CLI to Antigravity CLI is that the new one has improved support for multi-agent environments, but the company isn’t giving most of its users much of a choice on whether to switch. “On June 18, 2026, Gemini CLI and Gemini Code Assist IDE extensions will stop serving requests for Google AI Pro and Ultra, as well as those using it free of charge using Gemini Code Assist for individuals,” the Gemini CLI team wrote in their announcement of the transition. The change also affects Gemini Code Assist for GitHub, which won’t allow new installations beginning June 18, and will stop serving requests in the following weeks. Enterprises appear to be getting a pass, however, with Google noting that those using Gemini CLI or its IDE extensions through a Gemini Code Assist Standard or Enterprise license won’t see any changes in their access, nor will Gemini Code Assist for GitHub users accessing the tools through their enterprise Google Cloud accounts. “Gemini CLI will remain accessible via paid Gemini and Gemini Enterprise Agent Platform API keys,” Google explained. For everyone else, sorry: It’s Antigravity CLI or bust, but don’t expect the same experience. “There won't be 1:1 feature parity right out of the gate” between Gemini CLI and Antigravity CLI, Google added. Agent skills, hooks, subagents, and extensions are all being supported by Antigravity CLI at launch. But other stuff may take time to arrive, if it does at all. Pray we don’t alter the deal any further Take a look at the Gemini CLI GitHub page, and you'll find all the code that made it possible - it is an open-source project, after all. Surf over to the Antigravity CLI GitHub page and all you’ll see is a change log, readme, and a GIF file demonstrating the tool’s appearance. That’s right: Antigravity CLI isn’t open source - at least not from what Google has published so far - and it took developers no time at all to notice. Gemini CLI Lead Product Manager Dmitry Lyalin took to GitHub to make an announcement detailing some additional info about the forced CLI tool migration, and the comments are rife with people frustrated by the move. No small portion of the vitriol is targeted at apparent usage limits, with multiple people reporting they’d hit their weekly quota with just a couple of requests. The issues page for Antigravity CLI similarly has numerous posts asking Google to look into usage limits. Other posts accuse Google of using open-source contributions to improve a new closed-source product and generally express frustration with Google for killing yet another thing customers relied on. At the same time, Lyalin teased developers by telling them that, no, Gemini CLI isn’t truly gone if you’re willing to pay top dollar for it. “The project remains available to the community as an Apache 2.0 licensed repository with no changes,” Lyalin noted in his GitHub post. “You will continue to see us work on GitHub as we keep Gemini CLI updated with latest model releases, bugs and security fixes for our enterprise customers.” Now please open your wallets if you want access to this open-source product. ® ]]></content:encoded>
                <enclosure url="https://image.theregister.com/?imageId=259329&amp;width=800" type="image/jpeg" />
                <media:thumbnail url="https://image.theregister.com/?imageId=259329&amp;width=800" />
            </item>
    <item>
        <guid isPermaLink="true">https://www.theregister.com/a/5242899</guid>
        <link>https://www.theregister.com/ai-ml/2026/05/19/frustrated-franchisee-sues-pizza-hut-over-crappy-kitchen-ai/5242899</link>
        <pubDate>Tue, 19 May 2026 23:09:13 +0200</pubDate>
        <title>Frustrated franchisee sues Pizza Hut over crappy kitchen AI</title>
        <description><![CDATA[ The group claims $100M in losses from flubby system ]]></description>
        <category>ai + ml</category>
                <lab:kicker><![CDATA[ AI + ML ]]></lab:kicker>
                <dc:modified>Thu, 21 May 2026 12:56:51 +0000</dc:modified>
                <content:encoded><![CDATA[ The back-of-house AI system that Pizza Hut has mandated its restaurants to adopt has been so poorly received by some franchisees, that one is suing the company for $100 million in losses tied to the technology. Put that in your crust and stuff it! Chaac Pizza Northeast, a franchisee with around 111 Pizza Hut locations in New York, New Jersey, Maryland, Washington DC, and Pennsylvania, filed a complaint [PDF] in the Business Court of Texas earlier this month accusing the Hut of breaching its franchise agreement by mandating Chaac adopt restaurant management AI from Dragontail, a provider of AI-powered food delivery software. What was supposed to be a platform that would unify multiple kitchen systems under one AI-managed umbrella allegedly turned out to be a disaster for Chaac, which claims it was a leader among Pizza Hut franchises on metrics like delivery speed and rack time (i.e., the time between a pizza leaving the oven and leaving the store for delivery) prior to forced Dragontail adoption. Pizza Hut parent company Yum Brands purchased Dragontail in 2021. “With the intention to improve efficiency and service to the customer, Dragontail did the exact opposite; it caused significant delays and pummeled consumer satisfaction,” the lawsuit filing states. Chaac further alleged that Pizza Hut didn’t provide promised Dragontail support, and refused to allow Chaac to step back its use of the product, “causing cascading operational breakdowns and customer dissatisfaction.” Chaac admits it might be a bit of a special case, however, because of its particular business model: The company’s Pizza Hut locations don’t have a dining room, instead exclusively offering carry out and delivery services. Chaac also doesn’t employ its own drivers, instead relying on DoorDash to handle its deliveries. Before Dragontail’s implementation, staff at Chaac Pizza Huts had to input pickup requests into a DoorDash tablet, according to the lawsuit, which would handle getting the delivery order to a driver. Centralizing all of the order-to-delivery pipeline under one product meant that DoorDash gained visibility into the entire pizza making process. On one side that makes things more efficient, as the complaint explains. “This access allowed DoorDash to know when the pizzas went into the oven and were ready for pick-up, and when other pizza orders would be ready for pick-up,” the suit states - not bad if that means drivers aren’t sitting around waiting. In practice, however, that’s not what happened. Drivers were able to see whether additional orders would be up soon, meaning many of them would grab one order and simply wait 15 minutes for another, meaning the first order was invariably late and cold by the time it got to a customer. DoorDash drivers were also able to see any pre-paid tips on the order and whether an order was paid in cash. In many cases, drivers would decline tipless and cash orders. “These issues, arising out of DoorDash’s visibility, caused a disruption in orderly delivery and significantly slower delivery times,” the suit claimed, adding that the changes ultimately benefited DoorDash at Chaac’s expense. “The damage was not abstract,” the suit continued. “Chaac suffered lost revenue, lost profits, loss in enterprise value, business interruption, and erosion of goodwill and customer relationships” as a result of Dragontail adoption. According to the lawsuit, loss of business and enterprise value due to the forced adoption of kitchen management AI caused is in excess of $100 million, which Chaac is demanding as recompense. It’s not difficult to find examples online of Pizza Hut employees complaining about Dragontail. Multiple Reddit threads from inside the 2020-2024 implementation period contain examples of employees describing dissatisfaction with the software. Several commenters note, as Chaac did in its lawsuit, that Dragontail took control out of the hands of its kitchens and put it in the hands of AI. “Dragontail’s integration with kitchen workflow and aggregator dispatch predictably stripped Chaac’s managers of operational control, introduced delays, and invited stacking and other algorithmic behaviors that slowed production and delivery,” the lawsuit argues. Pizza Hut has been struggling in recent years, with Yum closing hundreds of locations so far this year in the midst of a turnaround effort that included initiatives like adding Dragontail to the struggling brand’s locations; the company didn’t respond to questions for this story. Whether this’ll be another nail in Pizza Hut’s coffin or just a bump in the road will be up to a judge to decide. ® ]]></content:encoded>
                <enclosure url="https://image.theregister.com/?imageId=5243006&amp;width=800" type="image/jpeg" />
                <media:thumbnail url="https://image.theregister.com/?imageId=5243006&amp;width=800" />
            </item>
    <item>
        <guid isPermaLink="true">https://www.theregister.com/a/5242888</guid>
        <link>https://www.theregister.com/software/2026/05/19/firefox-151-helps-you-edit-pdfs-and-switch-oses/5242888</link>
        <pubDate>Tue, 19 May 2026 20:20:00 +0200</pubDate>
        <title>Firefox 151 helps you edit PDFs – and switch OSes</title>
        <description><![CDATA[ Export a profile on Windows, restore it on Linux. Extensions and themes too ]]></description>
        <category>software</category>
                <lab:kicker><![CDATA[ Software ]]></lab:kicker>
                <dc:modified>Tue, 19 May 2026 18:05:29 +0000</dc:modified>
                <content:encoded><![CDATA[ Firefox version 151 is out of beta and trickling out to users, with handy additions, just in case you were thinking of jumping ship from Windows 11 to Linux. Mozilla has officially released Firefox 151, although automatic updates are not yet happening at the time we write this. Its profit-making subsidiary MZLA has also released Thunderbird 151, although its new-feature list has less cool new shiny. The Firefox product announcement trumpets a “fresh new look and feel” for the New Tab page. As we’ve already lightly customized ours, we didn’t see that, but you know how it is – this is the sort of thing marketing folks can understand and sound excited about. Apparently you can customize its wallpaper and add a “Recent Activity” feed, if that’s what you want. (We’ve just added a few more rows of shortcuts to recent pages.) A more useful function, especially if you don’t trust Firefox Sync and you’re thinking of changing to a new OS, is improved handling of Firefox Backup, the built-in tools for backing up and restoring your profile (or profiles, plural, for the truly hardcore). The page in the last link hasn’t changed in the last three weeks, and it still says, “Note: Firefox Backup is currently only available to users on Windows 10 and 11. This feature may be extended to other platforms in future versions of Firefox.” Well, now it has: the release notes say it works on Linux now. We’ve also seen reports that it is now on macOS too, but not on our iMac (This could be because we’ve been using Firefox Sync since the late lamented Xmarks shut down). A key addition is that a profile backed up on one OS can now be restored on a different OS, which sounds like a significant improvement to us. This includes extensions and themes. Last time around, we shared the news that the PDF editor could split multipage PDFs into chunks, including saving out individual pages. In this version, it can now merge multiple PDFs into one, which also sounds handy. It’s the sort of feature we rarely need, but when we do, we really need it. Suffice to say that with recent Firefox versions, we no longer need a standalone PDF viewer. As well as over 30 security fixes and the usual developer changes, this release fixes some more visible bugs: multi-monitor handling has been improved, as has macOS integration. For instance, it can now handle links pasted from iOS using Apple’s Universal Clipboard feature, and dropdown menus on web pages use the native Apple menu style. Firefox’s Enhanced Tracking Protection has been further – er – enhanced, and now conceals more info about you – and much more on macOS. Thunderbird 151 is nigh upon us The closest thing to a universal cross-platform messaging client that the 21st century has to offer us so far has been updated, too. Thunderbird 151 is rolling out, although we haven’t been offered the update yet. The release notes' What’s New section only has three bullet points, and one of those is for the not-yet-public Thundermail service, part of Thunderbird Pro. However, it’s easier to adjust authorization settings for automatically-created accounts, Microsoft Exchange handling has been slightly tweaked, and you can sort tasks by different criteria. Since our task list is about three pages long and never seems to get any shorter, that sounds quite handy. ® ]]></content:encoded>
                <enclosure url="https://image.theregister.com/?imageId=5242942&amp;width=800" type="image/jpeg" />
                <media:thumbnail url="https://image.theregister.com/?imageId=5242942&amp;width=800" />
            </item>
    <item>
        <guid isPermaLink="true">https://www.theregister.com/a/5242728</guid>
        <link>https://www.theregister.com/security/2026/05/19/drupal-warns-admins-to-brace-for-highly-critical-core-patch/5242728</link>
        <pubDate>Tue, 19 May 2026 17:56:25 +0200</pubDate>
        <title>Clear your calendar, Drupal user: You have a critically urgent patch to install</title>
        <description><![CDATA[ The org’s staying mum on the details, but Wednesday’s fixes reach back to unsupported 8.9 branches ]]></description>
        <category>security</category>
                <lab:kicker><![CDATA[ Security ]]></lab:kicker>
                <dc:modified>Wed, 20 May 2026 11:34:01 +0000</dc:modified>
                <content:encoded><![CDATA[ Updated: If you use Drupal, get ready to patch without delay. The org behind the popular open source content management system is warning of a highly critical vulnerability in Drupal core that is serious enough for it to tell users ahead of Wednesday’s patch release to set aside time to install the fix immediately. The Drupal Security Team’s Monday PSA announcing the imminent patch for Drupal core doesn’t include any specifics, with the PSA noting that Drupal isn’t willing to share additional information until the announcement is made alongside the patch release. That, says Drupal, will happen at some point between 1700 and 2100 UTC on Wednesday, May 20. To reiterate, this vulnerability is found in Drupal core, the bare-bones version of Drupal designed for developers, and not Drupal CMS, the preconfigured version for those who want Drupal but don’t have coding skills. Drupal noted that sites using Drupal Steward, its paid web application firewall service, are protected against known attack vectors, though it still recommends Steward customers update their core instances in case additional exploit methods emerge. “The Drupal Security Team urges you to reserve time for core updates at that time because exploits might be developed within hours or days,” the advisory warns. Drupal also recommends users update to the latest supported release prior to Wednesday’s patch “so that you can address any other upgrade issues before the security window." While it won’t get specific on the nature of the vulnerability, Drupal did share its severity score based on NIST’s standard scoring methodology, and it’s not good: The bug scored 20 out of a max of 25 on that scale, as defined by Drupal’s own documentation. More specifically, it’s trivially easy to leverage, doesn’t require any privilege level to exploit, could make all non-public data on an affected site accessible to the attacker, and could allow an attacker to modify or delete whatever they wanted. The only two things preventing it from scoring a perfect 25/25 are the fact that a known exploit doesn’t exist yet and that it doesn’t affect all configurations, only those using “uncommon module configurations.” Drupal noted that security releases will be published on Wednesday for all currently supported core branches (11.3.x, 11.2.x, 10.6.x, and 10.5.x), as well as unsupported Drupal 11.1.x and 10.4.x branches for sites that have not yet upgraded from older 10.x and 11.x releases. Drupal users on 8.9 and 9.5 are also getting patches “given the potential severity of this issue,” though the advisory warns 8.9 and 9.5 users will need to install those updates manually, which “might introduce other bugs or regressions,” leading Drupal to recommend a full upgrade to a supported core branch. “Drupal 8 and 9 include numerous other, previously disclosed, security vulnerabilities that will not be addressed by either Drupal Steward or the best-effort patch files,” the advisory said. Drupal 7 users are safe. Given the fact that not all Drupal core environments will be affected, the advisory recommends all Drupal core users set aside time on Wednesday to determine whether they’re part of the vulnerable class, and take action immediately if so. ® Updated to add on May 20: The Drupal Security Team has been in contact to warn that, while Core is the primarily vulnerable product, Core's inclusion in Drupal CMS means those environments might be vulnerable too, so anyone running Drupal will need to be sure their site is secure. As for the patch itself, Drupal told us it can be installed in "minutes or maybe seconds depending on the site," which likely won't need to be taken offline in order to install the patch.In other words, you really ought to be sure this gets installed before you're caught being a straggler. ]]></content:encoded>
                <enclosure url="https://image.theregister.com/?imageId=5242858&amp;width=800" type="image/jpeg" />
                <media:thumbnail url="https://image.theregister.com/?imageId=5242858&amp;width=800" />
            </item>
    <item>
        <guid isPermaLink="true">https://www.theregister.com/a/5240771</guid>
        <link>https://www.theregister.com/ai-ml/2026/05/14/ontario-auditors-find-doctors-ai-note-takers-routinely-blow-basic-facts/5240771</link>
        <pubDate>Thu, 14 May 2026 22:50:05 +0200</pubDate>
        <title>Sick and wrong: Ontario auditors find doctors' AI note takers routinely blow basic facts</title>
        <description><![CDATA[ 60% of evaluated AI Scribe systems mixed up prescribed drugs in patient notes, auditors say ]]></description>
        <category>ai + ml</category>
                <lab:kicker><![CDATA[ AI + ML ]]></lab:kicker>
                <content:encoded><![CDATA[ The AI systems approved for Ontario healthcare providers routinely missed critical details, inserted incorrect information, and hallucinated content that neither patients nor clinicians mentioned, according to a provincial audit of 20 approved vendors’ systems. The findings come from the Office of the Auditor General of Ontario, Canada, and are included in a larger report about the state of AI usage by public services in the province. They specifically address the AI Scribe program, the Ontario Ministry of Health initiated for physicians, nurse practitioners, and other healthcare professionals across the broader health sector. As part of the procurement process, officials conducted evaluations using simulated doctor-patient recordings. Medical professionals then reviewed the original recordings alongside the AI-generated notes to evaluate their accuracy. What they found was, frankly, shocking for anyone concerned about the accuracy of AI in critical situations. Nine out of 20 AI systems reportedly “fabricated information and made suggestions to patients' treatment plans” that weren’t discussed in the recordings. According to the report, evaluators spotted potentially devastating incorrect information in the sample reports, such as no masses being found, or patients being anxious, even though these things were never discussed in the recordings. Twelve of the 20 systems evaluated inserted incorrect drug information into patient notes, while 17 of the systems “missed key details about the patients’ mental health issues” that were discussed in the recordings. Six of the systems “missed the patients’ mental health issues fully or partially or were missing key details,” per the report. OntarioMD, a group that offers support for physicians in adopting new technologies and was involved in the AI Scribe procurement process, has recommended that doctors manually review their AI notes for accuracy, but the report notes there’s no mandatory attestation feature in any of the AI Scribe-approved systems. Bad evaluations don’t help, either AI systems making mistakes isn’t exactly shocking. As we’ve reported previously, consumer-focused AI has a tendency to provide bad medical information to users, and some studies have found large language models failed to produce appropriate differential diagnoses in roughly 80 percent of tested cases. But the tools evaluated here are for doctors, not consumers, and such poor performance necessitates explanation. A good portion of the report blames how the systems were evaluated. According to the report, the weight given to various categories of AI Scribe performances was wonky. While 30 percent of a platform’s evaluation score depended solely on whether they had a domestic presence in Ontario, the accuracy of medical notes contributed only 4 percent to the total score. Bias controls accounted for only 2 percent of the total evaluation score; threat, risk, and privacy assessments counted for another 2 percent; and SOC 2 Type 2 compliance contributed an additional 4 percentage points. In other words, criteria tied to accuracy, bias controls, and key security and privacy safeguards made up only a small portion of the total evaluation score for the AI Scribe systems. “Inaccurate weightings could result in the selection of vendors whose AI tools may produce inaccurate or biased medical records or lack adequate protection to safeguard sensitive personal health information,” the report said of the scoring regime. The Register reached out to the Ontario Health Ministry for its take on the report, and whether it was going to conform to its recommendations for the AI Scribe program, but we didn’t immediately hear back. A spokesperson for the Ministry told the CBC on Wednesday that more than 5,000 physicians in Ontario are participating in the AI Scribe program and there have been no known reports of patient harms associated with the technology. ® ]]></content:encoded>
                <enclosure url="https://image.theregister.com/?imageId=5240796&amp;width=800" type="image/jpeg" />
                <media:thumbnail url="https://image.theregister.com/?imageId=5240796&amp;width=800" />
            </item>
    <item>
        <guid isPermaLink="true">https://www.theregister.com/a/5240005</guid>
        <link>https://www.theregister.com/software/2026/05/13/googles-ai-enabled-mouse-pointer-understands-this-and-that/5240005</link>
        <pubDate>Thu, 14 May 2026 00:19:47 +0200</pubDate>
        <title>Google's AI-enabled mouse pointer understands 'this' and 'that'</title>
        <description><![CDATA[ Right-clicking could go the way of the 3.5-inch floppy at the Chocolate Factory ]]></description>
        <category>software</category>
                <lab:kicker><![CDATA[ software ]]></lab:kicker>
                <content:encoded><![CDATA[ Google doesn't design mouse traps, so it's trying to design a better mouse. Google DeepMind announced a research effort to transform the standard computer mouse cursor into a context-aware, AI-powered tool, marking what the company described as the first major rethinking of the cursor in more than 50 years. The project by researchers Adrien Baranes and Rob Marchant integrated Google's Gemini AI model with an experimental context-aware mouse pointer. In this way, the company said, the system can understand where a user clicks, what they are clicking on, and the likely intent behind the interaction. Researchers said there is a persistent friction in how people currently interact with AI tools. Most AI assistants today live in a separate window, requiring users to copy, paste, or drag content into a chat interface before receiving help. The new approach aims to reverse that dynamic. "We want the opposite: intuitive AI that meets users across all the tools they use, without interrupting their flow," the researchers stated in the blog post. The mouse pointer works alongside the computer’s microphone, allowing Gemini to listen as the user points. This lets users refer to features on the screen with object pronouns like “this” and “that.” In a demonstration website, a user can hover a cursor over a crab and say “move this here,” and the system understands enough context to grab the crab and move it to where the cursor indicates. The first computer mouse, a one-button prototype with metal wheels for the x- and y-axis, was built out of wood in 1964 and was patented in 1970 by its inventors Doug Engelbart and Bill English, who worked at the Stanford Research Institute. Engelbart foresaw a day when humans and computers would interact more easily and naturally, which he talked about during his 1997 acceptance speech for the Lemelson-MIT Prize. “The computer technology, the digital capabilities, it’s affecting communications, displays, storage, computer processing. It’s affecting the way you can interface to things a lot more flexibly,” he said. “That’s going to be so pervasively high-impact in our society and our organizations that it's more than anything we’ve had to cope with evolutionary wise.” Maintain the flow At Google, the team said it laid out four design principles guiding the project. The first, which the researchers called "Maintain the flow," stated that AI capabilities should work across all applications rather than forcing users into separate AI-specific environments. Under this principle, a user could point at a PDF and request a summary, or hover over a statistics table and ask for a chart, all without leaving the current application. The next, "Show and tell," addressed the burden of prompt writing. The researchers stated that an AI-enabled pointer could capture visual and semantic context from the screen, reducing the need for users to write detailed text instructions to the model. They also developed the AI cursor based on how humans naturally communicate using short phrases and gestures like “this” and “that.” The researchers stated that the system would allow users to issue commands like "Fix this" or "Move that here" while the AI fills in the contextual gaps. The fourth principle, "Turn pixels into actionable entities," lets the pointer recognize structured objects within on-screen content. The researchers stated that this capability could turn a photo of a handwritten note into an interactive to-do list, or convert a paused video frame showing a restaurant into a booking link. In the blog, the researchers said that Google DeepMind has already begun integrating the lessons learned into products. A feature called Magic Pointer will soon roll out on the forthcoming Googlebook laptop platform, which The Chocolate Factory introduced earlier this week. The company said the technology will also allow users of Gemini in Chrome to point at specific parts of a webpage and ask questions, rather than composing a full text prompt. Experimental demos of the AI-enabled pointer are currently available through Google AI Studio, where users can test image-editing and map-based interactions using the point-and-speak approach. The company said it plans to continue testing the concept across additional platforms, including Google Labs' Disco. ® ]]></content:encoded>
                <enclosure url="https://image.theregister.com/?imageId=5240038&amp;width=800" type="image/jpeg" />
                <media:thumbnail url="https://image.theregister.com/?imageId=5240038&amp;width=800" />
            </item>
    <item>
        <guid isPermaLink="true">https://www.theregister.com/a/5239967</guid>
        <link>https://www.theregister.com/ai-ml/2026/05/13/anthropic-butts-in-to-small-business-promises-help-with-payroll-and-other-core-tasks/5239967</link>
        <pubDate>Wed, 13 May 2026 22:48:46 +0200</pubDate>
        <title>Anthropic butts in to small business, promises help with payroll and other core tasks</title>
        <description><![CDATA[ But Pro or Max biz users should know that the company may train its AI on your data ]]></description>
        <category>ai + ml</category>
                <lab:kicker><![CDATA[ AI + ML ]]></lab:kicker>
                <content:encoded><![CDATA[ Anthropic is pushing into the small business space with a set of new plug-and-play tools designed for those without a tech team budget, but be warned: Depending on your Anthropic subscription tier, some business data might get sucked up to train Claude. Anthropic announced Claude for Small Business (CSB) on Wednesday, describing the new plugin as a way for SMB owners without AI expertise to automate the basic business tasks they’re saddled with, like payroll, chasing payments, and launching campaigns, that are usually the purview of different departments at the enterprise level. Installation is designed to be dead simple, with Pro, Max, and Teams plan users able to add it as a plugin from the Cowork space in the Claude Desktop app. Skills can then be run using natural language prompts or slash commands outlined here. Users will find “a package of connectors and ready-to-run workflows” inside the CSB plugin, according to the announcement. The aforementioned capabilities of the plugin are part of 15 skills based on common repeatable business tasks, while 15 agentic workflows are also included across areas like finance, operations, marketing and the like. As for the connectors themselves, Anthropic specifically mentions seven of them included in Claude for Small Business: Intuit Quickbooks, PayPal, HubSpot, Canva, Docusign, Google Workspace, and Microsoft 365. An Anthropic spokesperson told The Register in an email that CSB isn’t limited to those connectors, but the skills and workflows rolling out for the plugin were only optimized for those connectors to start. Anthropic told us it chose those products based on the results of a survey of SMB owners, but it plans to add support for more connectors in the coming months. In other words, if you’re a small business owner and you rely on a platform not on that list, you’ll have to keep waiting a while longer if you want to pull that info into Claude. Gotta reach ‘em all It's logical that Anthropic is pushing into the SMB space. The company has seen a leap in business customer subscriptions this year, taking advantage of OpenAI’s slip in the professional user space, and with growth comes the search for new markets to tap. As Anthropic notes in the announcement, and as many analysts have pointed out, AI adoption among SMBs has historically lagged enterprises. That’s to be expected, of course: Enterprises have far more resources to invest in new, unproven technologies and the money to absorb failure when said new tech doesn’t pan out as expected. Anthropic said in its CSB announcement that it specifically designed the new plugin for “those who have historically been last in line for new technology,” or small businesses, in other words. The company also launched an AI fluency for small business course to help SMB owners understand what exactly they’re installing when they tell Claude to install CSB. But if you're taking part, you have to be OK with the idea that Anthropic might train its AI on your business data. Anthropic points out in the announcement that it doesn't train its AI models on the data of its business customers “on our Team and Enterprise Plans.” But as we noted above, Anthropic is marketing CSB to those on Pro, Max, and Teams plans, and the privacy policy page for Pro and Max says something quite different: “We will use your chats and coding sessions (including to improve our models),” the page states. “Chat and coding session data we may use for improving our models includes the entire related conversation, along with any content, custom styles or conversation preferences, as well as data collected when using Claude for Chrome.” Raw content from connectors isn’t included, the page explains, “though data may be included if it’s directly copied into your conversation with Claude.” This only applies to users who, under regular circumstances, have chosen to allow Anthropic to use chats to improve Claude, but it likely won’t shock any El Reg readers to learn that permission is on by default – Anthropic told us that it's on users to turn it off. If you're copacetic with all this, you can start using CSB today - there’s no extra cost associated with installing the tool for anyone on a Pro, Max, or Teams plan. ® ]]></content:encoded>
                <enclosure url="https://image.theregister.com/?imageId=5223101&amp;width=800" type="image/jpeg" />
                <media:thumbnail url="https://image.theregister.com/?imageId=5223101&amp;width=800" />
            </item>
    <item>
        <guid isPermaLink="true">https://www.theregister.com/a/5238841</guid>
        <link>https://www.theregister.com/databases/2026/05/13/dbase-debased-database-titan-fades-to-black-after-47-years/5238841</link>
        <pubDate>Wed, 13 May 2026 12:44:00 +0200</pubDate>
        <title>dBase debased: Database titan fades to black after 47 years</title>
        <description><![CDATA[ Blog post mourning decline appears to have helped knock what was left of the veteran app's online presence offline ]]></description>
        <category>databases</category>
                <lab:kicker><![CDATA[ Databases ]]></lab:kicker>
                <dc:modified>Wed, 13 May 2026 13:59:30 +0000</dc:modified>
                <content:encoded><![CDATA[ It looks like a popular blog post about the decline and fall of dBase has knocked the long-moribund database's website offline. Sic transit gloria mundi? We were rather entertained by a recent blog post on "Delphi Nightmares" mourning the passing of the online store for the dBase website: dBase: 1979-2026. When the post went up, the online shop at store.dbase.com was still online, but since the post was shared on Hacker News yesterday, even that has gone. One could say that after 47 years, dBase has finally been debased. It's an interesting telling of the decline and fall of what was once an industry titan, and for us, the disappearance of the site itself once the blog post went up is just the cherry on top. Indirectly, what turned into dBase started out as a tool called JPLDIS, written for the Jet Propulsion Laboratory's three Univac 1108 computers. A FORTRAN rewrite of the simpler Tymshare RETRIEVE [PDF] tool, it was started by Jack Hatfield and finished by Jeb Long. C. Wayne Ratliff then rewrote it in Intel 8080 assembly language for PTSDOS on his IMSAI 8080, and tried to sell it under the name Vulcan: he put an advert in BYTE Magazine, offering it for $50. It wasn't a hit, as he recounted in an interview with Susan Lammers. Serial entrepeneur Ed Tate hired him and licensed Vulcan. Tate set up a new company called Ashton-Tate – there was no Ashton, but he later bought a parrot, named it Ashton and made it the mascot. Ashton-Tate renamed the database to dBASE II – to sound more mature – raised the price dramatically, and sold the CP/M version as shrink-wrap software.The late John Walker noted in 1982 that it was "selling like hotcakes at $800 a pop." That same year, a PC version of dBase II became one of early commercial business applications for IBM's new PC. Former dBase Developer's Bulletin editor Jean-Pierre Martel's personal history of dBASE recounts how it remained one of the industry-standard apps throughout the 1980s. In 1984, the enhanced dBase III did even better, followed in 1986 by dBase III+, with a menu-driven UI as well as the infamous "dot prompt" command line. In 1988, dBase IV followed, but didn't include the promised compiler for the dBase programming language. This opened up opportunities for rivals. Nantucket's Clipper was one, which could compile dBase code into applications. It was already out there: because it didn't include the interactive language, that meant it didn't have the same primary UI, which protected it from being sued. Clipper ended up acquired by Computer Associates. Fox Software's FoxBase, later FoxPro, was another, and even Ratliff himself was impressed. Microsoft eventually acquired FoxPro. There were many others, and that was the real problem for Ashton-Tate and the dBase product: its programming language became standardized, and because of trademark issues, known as xBase. Even before the era of "open source," there was a DOS shareware app called WAMPUM, which is still out there. There are a number of FOSS implementations, including Harbour and its fork xHarbour. The Harbour GitHub repo has seen some activity this year, and the xHarbour one some too. Once your expensive proprietary app's file format and programming language escape into the wild and become partially standardized, that can make it hard to keep making money from it. It looks like that finally spelled the end for dBase LLC… but in the meantime, the xBase language is alive and reasonably well considering its advanced age for a bit of software. ® ]]></content:encoded>
                <enclosure url="https://image.theregister.com/?imageId=5238863&amp;width=800" type="image/jpeg" />
                <media:thumbnail url="https://image.theregister.com/?imageId=5238863&amp;width=800" />
            </item>
    <item>
        <guid isPermaLink="true">https://www.theregister.com/a/5238623</guid>
        <link>https://www.theregister.com/software/2026/05/12/eu-law-bestows-6m-more-firefox-users-upon-us-moz-says/5238623</link>
        <pubDate>Tue, 12 May 2026 13:29:42 +0200</pubDate>
        <title>EU browser choice rules send millions more users Firefox's way</title>
        <description><![CDATA[ Mozilla claims the Digital Markets Act delivered lasting bump, invites Britain to do similar ]]></description>
        <category>software</category>
                <lab:kicker><![CDATA[ Software ]]></lab:kicker>
                <dc:modified>Tue, 12 May 2026 11:30:50 +0000</dc:modified>
                <content:encoded><![CDATA[ The EU's Digital Markets Act (DMA) has been kind to Mozilla, which says Firefox use is on the up as Europeans are given a choice of default browser on mobile. Through these browser selection screens, the company reckons 6 million users have opted for Firefox instead of what would otherwise have been Safari or Chrome, depending on whether they used an iPhone or Android device. Moz has seen the greatest success on iGadgets, with a 113 percent increase compared to a mere 12 percent rise on Android. This is less likely to be explained by overwhelming disdain for Safari than by the ways in which Apple and Google implemented these browser choice screens. Android devices display the browser selection screens upon first boot or after factory reset, whereas iPhone and iPad users are now shown the same screen as soon as they open Safari for the first time. The DMA obligations began applying in March 2024. Apple's implementation of the EU requirements was always going to lead to more people being prompted to select their browser than Google's, which mostly applies to new Android owners after the DMA was enforced, rather than existing users. Mozilla won't care, though, because not only are user numbers up, but user retention is also looking good – it is five times higher than before the DMA, by its reckoning. Other browser vendors have reported similar results, according to a recent European Commission review [PDF] of the DMA's efficacy, although it didn't cite any specific figures. Few vendors have published long-term results like Mozilla's, although Aloha, Brave, Opera, and Vivaldi all reported sizable uplifts in users in the initial days and weeks following the DMA's enforcement. Further, in recent publications [PDF], DuckDuckGo said around 40 percent more users selected its browser on Android thanks to the DMA browser choice screen. The privacy-focused tech biz offered the statistic in its submission to the UK government's consultation on how to maintain competition in online search. Moz also submitted its thoughts on the topic, and unsurprisingly, given they both benefited massively from them, both vendors want the same DMA-style browser choice screens to feature in the UK market. DuckDuckGo said they should be shown to users annually, and Google should be forced to remove its "Switch back to Google" prompt in Chrome. Mozilla wants the browser choice screens to be delivered to UK users in 2026, for the same users also to be presented with similar screens for default search engines, and for these measures to be enforceable rather than relying only on voluntary commitments from the relevant vendors. Criticizing the DMA, Moz added that it would also like to see the same measures applied to desktop browsers, alleging that Microsoft deploys deceptive design tactics to push its Edge browser. ® ]]></content:encoded>
                <enclosure url="https://image.theregister.com/?imageId=5225400&amp;width=800" type="image/jpeg" />
                <media:thumbnail url="https://image.theregister.com/?imageId=5225400&amp;width=800" />
            </item>
    <item>
        <guid isPermaLink="true">https://www.theregister.com/a/5238299</guid>
        <link>https://www.theregister.com/software/2026/05/11/ratty-terminal-emulator-brings-3d-graphics-to-the-command-line/5238299</link>
        <pubDate>Mon, 11 May 2026 22:54:30 +0200</pubDate>
        <title>Rodent-obsessed developer creates Ratty to bring 3D graphics to the command line</title>
        <description><![CDATA[ Inspired by TempleOS, this terminal emulator is just about as bonkers ]]></description>
        <category>software</category>
                <lab:kicker><![CDATA[ Software ]]></lab:kicker>
                <dc:modified>Mon, 11 May 2026 21:44:37 +0000</dc:modified>
                <content:encoded><![CDATA[ When you think of a terminal emulator, you imagine a command line interface filled with ASCII text and a prompt. However, one developer has reimagined the experience to include inline 3D objects and image support. Dubbed Ratty by its creator Orhun Parmaksiz for its 3D spinning rat cursor, the terminal window itself is a 3D canvas that supports sprites and 3D models, can render 3D drawings in real time, and even includes its own graphics protocol. “Terminal emulators are a big part of our daily lives as developers but yet we are not making enough innovations in that space,” Parmaksiz told The Register in an email. “With Ratty I hope to inspire others to experiment with terminals and push the limits of what they can do.” Parmaksiz wrote in his blog post introducing Ratty that he accomplished the whole thing using his own Rust terminal interface library, Ratatui, along with the Bevy game engine, also built with Rust. The aforementioned Ratty Graphics Protocol was created in order to register 3D assets and place them in an anchored terminal cell space. “Ratty separates terminal emulation from presentation: one side handles PTY I/O and terminal parsing, while the other turns the result into a GPU-rendered 2D or 3D scene,” Parmaksiz explained. “This allows for a lot of flexibility in how the terminal output is displayed (e.g. you can warp the whole damn thing).” Ratatui ends up serving as the terminal rendering layer, Parmaksiz explained, taking whatever the terminal state is, rebuilding it in its own buffer, and rendering said buffer onto a texture that is then rendered via Bevy. Given its design, be forewarned if you try to install and run Ratty: It’s going to eat up a lot of memory since it’s running a game engine. “I know, sacrificing 300 MB of RAM just to run a terminal emulator is a lot,” Parmaksiz said. “But everything comes with a cost, especially the spinning rat cursor.” Building the fourth temple Parmaksiz’s desire to push the limits of terminal emulators past their logical limits didn’t come from nowhere - he actually got inspiration from a source that some grey-hairs in the tech community might have been reminded of at the very beginning of this story: TempleOS. For those unfamiliar with TempleOS, it’s an operating system that was developed by the late Terry Davis, a schizophrenic, and arguably genius, software developer who believed he was building the OS at the command of God to serve as a digital Jewish Third Temple. Using TempleOS is an exercise in frustration given its confusing interface, not to mention deliberate constraints (Davis believed its 640x480 desktop, 16-color display, single-voice audio and other features were part of God’s commandment), but it also included a fascinating capability not seen in other OSes: first-class, insertable sprites on the command line. “I was blown away by the creativity and passion behind it,” Parmaksiz told us of TempleOS, noting that 3D command line sprites in the OS were his inspiration for Ratty. “I wanted to see how adopting that to a modern-day terminal emulator would look like and experimented with a couple of other things while I was at it. I'm super happy with the result!” Parmaksiz told us that a number of people instantly caught on to the TempleOS inspiration, and that the feedback has been overwhelmingly positive. That said, he also admitted that most people who’ve used it have been scratching their heads over an actual use case. “I think this will also clarify itself if we give it more time,” Parmaksiz said in his email. “I mean... I really would like to see a full-fledged CAD program in the terminal built with Ratty Graphical Protocol at some point!” Whether that’ll ever happen remains to be seen - this is purely a fun project for now and Parmaksiz isn’t even sure it’s in his personal time budget to continue to maintain. “I'm just testing the waters for now, but the reception has been amazing so far. I would be happy to continue development if people start using Ratty and start developing cool things with it,” Parmaksiz said, noting that the code is open and he’d be thrilled if others contributed. Parmaksiz has developed a Ratatui widget that enables devs to build applications that run in Ratty, like a temple runner knockoff. “My ultimate goal with Ratty is to explore the possibilities of what a terminal can be and inspire new ideas and projects in the terminal space,” Parmaksiz wrote in his blog post. “I believe these kinds of experiments are where creativity is born and I hope to spark some ideas for the future of terminals.” ® ]]></content:encoded>
                <enclosure url="https://image.theregister.com/?imageId=5238345&amp;width=800" type="image/jpeg" />
                <media:thumbnail url="https://image.theregister.com/?imageId=5238345&amp;width=800" />
            </item>
    <item>
        <guid isPermaLink="true">https://www.theregister.com/a/5238111</guid>
        <link>https://www.theregister.com/security/2026/05/11/anthropics-bug-hunting-mythos-was-greatest-marketing-stunt-ever-says-curl-creator/5238111</link>
        <pubDate>Mon, 11 May 2026 18:30:53 +0200</pubDate>
        <title>Anthropic’s bug-hunting Mythos was greatest marketing stunt ever, says cURL creator</title>
        <description><![CDATA[ After all that hype, AI scanner found one low-severity cURL flaw ]]></description>
        <category>security</category>
                <lab:kicker><![CDATA[ Security ]]></lab:kicker>
                <content:encoded><![CDATA[ cURL developer Daniel Stenberg has seen Anthropic’s Mythos, a model the AI biz has suggested is too capable at finding security holes to release publicly, scan his popular open source project. But after the system turned up just a single vulnerability, he concluded the hype around Mythos was “primarily marketing” rather than a major AI security breakthrough. Stenberg explained in a Monday blog post that he was promised access to Anthropic’s Mythos model - sort of - through the AI biz’s Project Glasswing program. Part of Glasswing involves giving high-profile open source projects access via the Linux Foundation, but while Stenberg signed up to try Mythos, he said he never actually received direct access to the model. Instead, someone else with access ran Mythos against curl’s codebase and later sent him a report. “It’s not that I would have a lot of time to explore lots of different prompts and doing deep dive adventures anyway,” Stenberg explained. “Getting the tool to generate a first proper scan and analysis would be great, whoever did it.” That scan, which analyzed curl’s git repository at a recent master-branch commit, was sent back to him earlier this month, and it found just five things that it claimed were “confirmed security vulnerabilities” in cURL. Saying he had expected an extensive list of vulnerabilities, Stenberg wrote that the report “felt like nothing,” and that feeling was further validated by a review of Mythos’ findings. “Once my curl security team fellows and I had poked on this short list for a number of hours and dug into the details, we had trimmed the list down and were left with one confirmed vulnerability,” Stenberg said, bringing us back to the aforementioned number. As for the other four, three turned out to be false positives that pointed out cURL shortcomings already noted in API documentation, while the team deemed the fourth to be just a simple bug. “The single confirmed vulnerability is going to end up a severity low CVE planned to get published in sync with our pending next curl release 8.21.0 in late June,” the cURL meister noted. “The flaw is not going to make anyone grasp for breath.” That said, Mythos did find several other non-security bugs that Stenberg said the team is working on fixing, and he notes that their description and explanation were well done. Mythos can do good work, in other words, but it’s not a ground-breaking, game-changing AI model like Anthropic has claimed. “My personal conclusion can however not end up with anything else than that the big hype around this model so far was primarily marketing,” Stenberg said in the blog post. “I see no evidence that this setup finds issues to any particular higher or more advanced degree than the other tools have done before Mythos.” cURL code is no stranger to AI To say cURL has become widely used in its nearly three decades of existence would be an understatement. Its wide reach has meant that its team has been running it through all sorts of static code analyzers and fuzz testing it since well before the dawn of the AI age. With AI’s rise, the cURL team has adapted, meaning Mythos is hardly the first AI to get its fingers on cURL’s codebase. “These tools and the analyses they have done have triggered somewhere between two and three hundred bugfixes merged in curl through-out the recent 8-10 months or so,” Stenberg said of tools like AISLE, Zeropath, and OpenAI Codex Security that’ve tested cURL code. “A bunch of the findings these AI tools reported were confirmed vulnerabilities and have been published as CVEs. Probably a dozen or more.” Stenberg’s experience with AI testing cURL, in other words, makes it a great candidate to see how effective Mythos can really be at finding more than the average AI. As Stenberg noted elsewhere in his blog post, Mythos isn’t doing anything particularly novel when it comes to security discoveries: It might be a bit better at finding things than previous models, but “it is not better to a degree that seems to make a significant dent in code analyzing,” the cURL author noted. Stenberg isn’t an AI doomer when it comes to its ability to improve software design, though. Yes, he may have closed the cURL bug bounty earlier this year due to an influx of sloppy, useless bug reports, but he also noted a few months prior to the bounty closure that some security researchers assisted by AI have made valuable reports. “AI powered code analyzers are significantly better at finding security flaws and mistakes in source code than any traditional code analyzers did in the past,” Stenberg said, adding an important qualifier for the Mythos moment: “All modern AI models are good at this now.” Mythos isn’t any more creative than its creators Both older AI models and security-focused tools like Mythos have a common limitation, as far as Stenberg is concerned: They’re only as good at finding security vulnerabilities as the humans who programmed them. “AI tools find the usual and established kind of errors we already know about. It just finds new instances of them,” Stenberg said. “We have not seen any AI so far report a vulnerability that would somehow be of a novel kind or something totally new.” As for Mythos, Stenberg remains unimpressed, calling it "an amazingly successful marketing stunt for sure" in his blog post. In an email to The Register, Stenberg admitted that it’d be possible for AI models to actually discover new, novel types of vulnerabilities, but he’s still not convinced that they can go beyond what humans are capable of finding, given that they’re limited by our understanding of how software vulnerabilities work. At the end of the day, Stenberg explained, when we talk about security, we’re only talking about code. “Source code is text and it feels like maybe we already know about most ways we can do security problems in it,” he pondered in his email. In other words, like the valuable AI-assisted reports made to the cURL bug bounty program before its closure due to a flood of AI garbage, making valuable use of systems like Mythos is going to require humans to get creative. Sorry, no foisting your critical thinking onto a bot. “Human researchers have always used tools when they look for security problems,” Stenberg told us. “Adding AIs to the mix gives the humans even more powerful tools to use, more ways to find problems. I expect that many security bugs going forward will be found by humans coming up with new ways and angles of prompting the AIs.” Stenberg said that he hopes he’ll actually get his hands on Mythos so he can experiment with its capabilities, but he doesn’t seem to be holding out hope the promised access will materialize. “I have been promised access and for all I know I will eventually get it,” Stenberg told us. “I just don't know when.” ® ]]></content:encoded>
                <enclosure url="https://image.theregister.com/?imageId=5238164&amp;width=800" type="image/jpeg" />
                <media:thumbnail url="https://image.theregister.com/?imageId=5238164&amp;width=800" />
            </item>
    <item>
        <guid isPermaLink="true">https://www.theregister.com/a/5237702</guid>
        <link>https://www.theregister.com/personal-tech/2026/05/11/classic-outlooks-quick-steps-trip-over-microsoft-bug/5237702</link>
        <pubDate>Mon, 11 May 2026 12:28:11 +0200</pubDate>
        <title>Classic Outlook's Quick Steps trip over Microsoft bug</title>
        <description><![CDATA[ Client's handy automations get grayed out unless you know the keyboard shortcut ]]></description>
        <category>personal tech</category>
                <lab:kicker><![CDATA[ Personal Tech ]]></lab:kicker>
                <dc:modified>Mon, 11 May 2026 10:47:56 +0000</dc:modified>
                <content:encoded><![CDATA[ If you're using Quick Steps in Microsoft Outlook and wondering why they're grayed out, a bug introduced in version 2512 is the culprit. Classic Outlook is approaching the twilight years of its prodigiously long life, but users can still fall victim to productivity-killing bugs – in this case, a problem with Quick Steps. Quick Steps automates common or repetitive tasks in Outlook. Always have to move a bunch of messages to a specific folder? Quick Steps is your friend. Pin an email and mark it as unread? Again, the actions can be lined up in Quick Steps and executed with a single click or a keyboard shortcut. Until Microsoft breaks it. In a support article, Microsoft has confirmed that in some situations, Quick Steps in classic Outlook can appear grayed out. The workaround (if rolling back or switching clients isn't an option) is to use a keyboard shortcut. "The shortcut will work even if the Quick Step is grayed out in the user interface," Microsoft wrote. The problem is that if a Quick Step contains actions that "can't be fulfilled," it's grayed out. Microsoft's own the example states: "A Quick Step that moves a message to a folder and clears categories will be grayed out in messages where there are no categories applied." "This is known to happen with Quick Steps with Flags and Categories actions such as 'Clear flags on message' or 'Clear categories'." Classic Outlook has suffered several glitches of late. Microsoft admitted in April that it could occasionally chow down on system resources for no obvious reason. Then there was its tendency to explode when opening too many emails. Microsoft has been clear that Classic Outlook's days are numbered. Outlook 2024 is due to drop out of mainstream support in 2029. However, there remains much that Classic Outlook does which New Outlook doesn't, such as COM support. And, when Microsoft hasn’t broken them, Quick Steps. ® ]]></content:encoded>
                <enclosure url="https://image.theregister.com/?imageId=5237731&amp;width=800" type="image/jpeg" />
                <media:thumbnail url="https://image.theregister.com/?imageId=5237731&amp;width=800" />
            </item>
    <item>
        <guid isPermaLink="true">https://www.theregister.com/a/5234824</guid>
        <link>https://www.theregister.com/software/2026/05/09/macos-27-threatens-to-bury-time-capsule-foss-brings-a-shovel/5234824</link>
        <pubDate>Sat, 09 May 2026 14:25:00 +0200</pubDate>
        <title>macOS 27 threatens to bury Time Capsule, FOSS brings a shovel</title>
        <description><![CDATA[ Apple's old backup boxes only speak AFP and SMB1, but NetBSD under the hood gives them one last shot ]]></description>
        <category>software</category>
                <lab:kicker><![CDATA[ Software ]]></lab:kicker>
                <dc:modified>Fri, 08 May 2026 15:32:56 +0000</dc:modified>
                <content:encoded><![CDATA[ The next major release of macOS looks likely to remove Apple Filing Protocol (AFP) support, stopping Time Capsules from working… but life FOSS, uh, finds a way. The current version of macOS "Tahoe" 26.4 already has network Time Machine issues, especially for folks using Apple Time Capsules. It looks like macOS 27 may completely remove the network protocol they need. However, the Time Capsules run NetBSD under the hood, and that means that the FOSS world has been able to come up with a workaround. It's called TimeCapsuleSMB, and it aims to keep older Time Capsules usable with modern macOS. It's eight months since Apple released macOS 26, and the company's annual release schedule means that macOS 27 is looming. Although Cupertino hasn't told the world much about it yet, it is warning sysadmins to "prepare your network environment for stricter security requirements." Reading the bulletin, we found it rather clixby: while it firmly warns that security checks will become stricter, it doesn't spell out what products will change or how. Happily, there are elder Mac gurus out there who interpret Apple's sometimes Delphic utterances, and Howard Oakley is one of the greatest. In a post about networking changes coming in macOS 27, he translates that it will require TLS 1.2 or above. (The Register explained TLS back in 2002, and version 1.2 appeared about six years later.) However, he also warns that it could mean the end of AFP, which is basically Appletalk-over-TCP/IP version 3.4. AppleTalk was the Mac network protocol for file sharing from System 6 onward. In 2013, OS X 10.9 "Mavericks" made Microsoft's SMB the default file-sharing protocol in place of AFP, and it looks like AFP now faces the ax: it was officially deprecated in macOS 15.5. To be fair, macOS 26 Macs started displaying a warning to Time Capsule users nearly a year ago. Apple introduced the first model of Time Capsule in 2008, and the fifth-generation version in 2013. The company discontinued the whole AirPort product line in 2018. All generations only support AFP and SMB version 1. That’s the original version that appeared with LAN Manager in 1987, and we reported on Samba dropping SMB1 back in 2022. The good news is that even if Apple kills its original file-sharing protocol next year, the FOSS community is on the case and won't let working kit die. The Time Capsule hardware is essentially a box containing a Wi-Fi access point and a hard disk, and an Arm chip with just enough software to share that HDD as network-attached storage. Apple didn't write this software from scratch: it picked up and customized NetBSD for the job. The first four generations of Time Capsule (flat square boxes) run NetBSD 4, and the fifth-gen devices – the tall tower-shaped models from 2013 onward – run NetBSD 6. That gave Microsoft's James Chang an opening. Since the devices run NetBSD, it's possible to compile a newer version of Samba, and copy it somewhere that the tiny embedded Arm computer can find it. Teaching such old kit a new trick is never that easy, though, and he faced a number of challenges, which he details in the design section of the project README. Among them are machines that only have about 900 KB of available disk space – less than 1 MB – and a tiny 16 MB RAMdisk. He settled on Samba 4.8, which dates back to 2018, the same year Apple discontinued the product line, but which includes the necessary Time Machine support, via a module named vfs_fruit. The TimeCapsuleSMB docs are worth a read. We found his descriptions of how he worked around the hardware's very significant limitations impressive. Notably, on the early models, you'll need to manually reload the software every time you reboot the Time Capsule. The final model can do this automatically. Don't fret at the thought of backing up to such an elderly spinning hard disk: iFixit has descriptions of how to replace the drive in both the early models and the later ones too. ® ]]></content:encoded>
                <enclosure url="https://image.theregister.com/?imageId=5234894&amp;width=800" type="image/jpeg" />
                <media:thumbnail url="https://image.theregister.com/?imageId=5234894&amp;width=800" />
            </item>
    <item>
        <guid isPermaLink="true">https://www.theregister.com/a/5234541</guid>
        <link>https://www.theregister.com/software/2026/05/07/nhs-code-clampdown-draws-open-source-backlash/5234541</link>
        <pubDate>Thu, 07 May 2026 12:39:32 +0200</pubDate>
        <title>NHS code clampdown draws open source backlash</title>
        <description><![CDATA[ Plus a petition for the UK Civil Service to go FOSS by default ]]></description>
        <category>software</category>
                <lab:kicker><![CDATA[ Software ]]></lab:kicker>
                <content:encoded><![CDATA[ Negative reactions are mounting against the UK National Health Service's plan to back away from open source – and you can add your voice. On Monday, The Register reported that the management at the NHS told its tech leadership to wall off the organization's FOSS repositories due to concerns about new LLM bug-hunting tools finding security vulnerabilities. If you will pardon a Douglas Adams quotation, this has made a lot of people very angry and been widely regarded as a bad move. One of the first reactions that The Reg FOSS desk received was from the Free Software Foundation Europe (FSFE), which sent it to us both by email and direct message. The FSFE says NHS England should not hide public code behind closed doors, and we feel that it has a good point. If you agree, there's an open letter to which you can attach your name. It's called "An open letter asking NHS England to keep its code open" on the simple and memorable domain keepthingsopen.com. At the time of writing, it has 812 signatures. By the time you read this, this vulture, whose shattered limbs have been reassembled by the NHS more than once, will appear on the list too. As a more general point, there is also a petition to the UK Parliament: "Migrate UK civil service to open-source software for data sovereignty & security." As a sensible step toward digital sovereignty and independence from systems and services run by other countries – countries that may not always be the friendly allies they have been – this, too, strikes us as a good move. If public money is paying for computer software, the code should be public as well. ® ]]></content:encoded>
                <enclosure url="https://image.theregister.com/?imageId=5234606&amp;width=800" type="image/jpeg" />
                <media:thumbnail url="https://image.theregister.com/?imageId=5234606&amp;width=800" />
            </item>
    <item>
        <guid isPermaLink="true">https://www.theregister.com/a/5230420</guid>
        <link>https://www.theregister.com/software/2026/05/06/firefox-integrates-an-ad-blocker-but-not-to-block-ads/5230420</link>
        <pubDate>Wed, 06 May 2026 13:03:30 +0200</pubDate>
        <title>Firefox integrates an ad-blocker, but not to block ads</title>
        <description><![CDATA[ It's in Waterfox too, and there it does what you'd expect ]]></description>
        <category>software</category>
                <lab:kicker><![CDATA[ Software ]]></lab:kicker>
                <dc:modified>Wed, 06 May 2026 17:06:40 +0000</dc:modified>
                <content:encoded><![CDATA[ Firefox 149 quietly shipped an interesting new feature buried in the code. As Mozilla bug #2013888 documents, the browser maker incorporated Brave's Rust-based adblock engine back in March - a detail surfaced in a blog post by Shivan Kaul Sahib, VP of Privacy and Security at Brave. The important thing here is that although Firefox has picked up the core Rust code that Brave uses for its internal ad-blocker, that's not what Firefox is using it for. Right now, in fact, it's disabled by default, but as a post from the official Firefox Reddit account says: In other words, inclusion of the code is experimental, and it's not intended for blocking ads. That's presumably also why it didn't appear in the release notes for either the early March beta or the late March release. So, yes, there is code for an ad-blocker in the last two versions of Firefox, but it’s off by default, and there’s no user interface to enable it. (There are ways round this, and we'll return to that later.) That said, the code from Brave can do this – because as it happens, the privacy-enhanced Waterfox fork of Mozilla's browser is also experimenting with a built-in ad-blocker, and it's using the same code. Waterfox recently celebrated its 15th birthday, and recent releases have an experimental built-in ad-blocker. At the time of writing, the latest version is 6.6.12, and that version's release notes mention the experimental ad-blocker, and link to the feature's feedback page which has more info. This says: Waterfox recently integrated another popular add-on. The last major release, Waterfox 6.6.0 in August last year - rebased on the new Firefox 140 ESR - included a native vertical tab bar. As revealed in a 2024 blog post, this is based on an integrated version of the popular Tree Style Tab extension. The new ad-blocker works similarly. If you enable it, on restart, Waterfox looks for other ad-blockers, such as Reg FOSS desk recommendation uBlock Origin, and offers to disable them. We tried this, and the uBO icon in the toolbar is replaced by a no-entry symbol, with a small number overlaid to show how many ads were blocked on the current page. (For instance, just two for Astronomy Picture of the Day, but 38 – and counting – for MSN.co.uk. And that's on top of network-level blocking from our Pi-hole.) So, experimental or not, adblock-rs is in there and it does work. It is possible to enable the version embedded in the desktop version of Firefox. It's controlled by two settings in about:config. The problem is that enabling it is not as simple as installing an extension, because extensions are not allowed to change those advanced settings. There is an experimental add-on called adblock-rust Manager. The README there carefully walks you through the manual steps you need to perform to enable the engine and tell it what to block, and then the add-on can monitor it and tell you what it's doing. We tried it, and it seems to do a perfectly acceptable job. For now, though, unless you're curious, we suggest staying with uBlock Origin, which works fine and isn't going anywhere. ® ]]></content:encoded>
                <enclosure url="https://image.theregister.com/?imageId=5230481&amp;width=800" type="image/jpeg" />
                <media:thumbnail url="https://image.theregister.com/?imageId=5230481&amp;width=800" />
            </item>
    <item>
        <guid isPermaLink="true">https://www.theregister.com/a/5226088</guid>
        <link>https://www.theregister.com/software/2026/05/05/openai-exec-says-it-will-burn-50b-on-compute-this-year/5226088</link>
        <pubDate>Tue, 05 May 2026 23:02:01 +0200</pubDate>
        <title>OpenAI exec says company hopes to burn $50B of somebody else's money on compute this year</title>
        <description><![CDATA[ If the numbers are large enough, perhaps we won't question the math ]]></description>
        <category>software</category>
                <lab:kicker><![CDATA[ AI + ML ]]></lab:kicker>
                <dc:modified>Tue, 05 May 2026 21:01:41 +0000</dc:modified>
                <content:encoded><![CDATA[ An executive for ChatGPT maker OpenAI said in court testimony on Tuesday that the AI model developer expects to burn $50 billion on computing power before the end of the year. Cofounder and president Greg Brockman threw out the number, which was previously reported by Bloomberg, during OpenAI's closely watched legal battle with hype-fiend Elon Musk. If it wasn't obvious, that would be $50 billion of someone else's money. Nearly four years after ChatGPT kicked off the AI boom, OpenAI's leadership hasn't yet figured out how to turn a profit. Heck, the company can't even manage to hit its own revenue targets, if recent reports are to be believed. That hasn't stopped CEO Sam Altman from talking the likes of Microsoft, Amazon, SoftBank, Nvidia and others into issuing press releases claiming plans to invest tens of billions of dollars into his quest for AGI. (Or was it AI superintelligence? The goalposts haven't exactly been fixed in concrete.) We're not sure if "investment" adequately captures the roundabout financial engineering that's gone into these highly publicized deals. Many are contingent on OpenAI using some of the pledged cash to lease massive quantities of compute either directly from its backers or their partners. Back in February, Amazon, Nvidia, and SoftBank announced a $110 billion investment in the AI startup, at least $80 billion of which came with strings attached. For example, OpenAI would need to rent two gigawatts of Amazon's Trainium AI accelerators and deploy its top GPT models in AWS to claim $35 billion of the $50 billion promised by the cloud titan. Similarly, Nvidia's $30 billion investment was tied to the deployment of five gigawatts of training and inference compute capacity at an estimated cost of $300 billion. In other words, these companies' investments in OpenAI are really more of a discount or rebate. It raises the question: Can OpenAI actually burn $50 billion in 2026, or is it simply throwing out more big numbers in hopes of maintaining an air of unassailable momentum? Place your bets – ideally in the form of burning as many heavily subsidized tokens as you can now, before prices inevitably rise. We've reached out to OpenAI for comment; we'll let you know if we hear anything back. ® ]]></content:encoded>
                <enclosure url="https://image.theregister.com/?imageId=222229&amp;width=800" type="image/jpeg" />
                <media:thumbnail url="https://image.theregister.com/?imageId=222229&amp;width=800" />
            </item>
    <item>
        <guid isPermaLink="true">https://www.theregister.com/a/5219859</guid>
        <link>https://www.theregister.com/software/2026/05/05/astera-speaks-softly-and-carries-a-big-switch/5219859</link>
        <pubDate>Tue, 05 May 2026 22:05:07 +0200</pubDate>
        <title>Astera speaks softly and carries a big switch</title>
        <description><![CDATA[ High-speed connectivity without NVLink baggage ]]></description>
        <category>software</category>
                <lab:kicker><![CDATA[ AI + ML ]]></lab:kicker>
                <dc:modified>Tue, 05 May 2026 19:26:19 +0000</dc:modified>
                <content:encoded><![CDATA[ Astera Labs unveiled an alternative to Nvidia's NVSwitch for building rack-scale AI systems on Tuesday, claiming it will work with nearly any accelerator. The AI fabric switch, codenamed Scorpio X, crams 320 lanes of PCIe 6.0 connectivity into a single ASIC with 5.12 TB/s of bidirectional bandwidth. Historically, PCIe switches have been used in a variety of applications including scale-out compute fabrics. CPUs alone either didn't offer enough or fast enough lanes for all the GPUs, NICs, and storage required. So, rather than hanging everything off the CPU, a PCIe switch, often built into the NIC, was used to connect everything together. Astera contends that with a big enough switch, PCIe is a viable alternative to interconnects like NVLink, in the scale-up fabrics used to make dozens or more GPUs behave more like a single large one without needing to redesign their accelerators. However, Astera hasn't just built a bigger PCIe switch. Scorpio is equipped with many of the same in-network compute capabilities as Nvidia's NVSwitch, which help to accelerate collective communications. These communications are especially important for generative AI inference. Large language models have become rather chatty from a network standpoint as mixture-of-experts (MoE) architectures have caught on. MoE models are composed of multiple sub-models called experts. For each token generated, a different selection of experts, potentially running on different GPUs, may be used.  By moving collective communications to the switch, the GPUs spend less time waiting for the network to catch up and more time churning out tokens. Astera has gone so far as to develop a multicast operation optimized for MoE inference that it calls Hypercast. "One of the limitations of the standard multicast is the number of groups you can actually support, as well as the dynamic nature of needing to change those groups on the fly for mixture-of-experts models," Ahmad Danesh, AVP of product management at Astera, told El Reg. Where Scorpio fits in the scale-up ecosystem While there are clear benefits to using PCIe as a chip-to-chip interconnect, Scorpio isn't exactly a replacement for Nvidia's NVSwitch chips. NVSwitch 6, announced at CES in January, offers nearly 3x the bandwidth at 14.4 TB/s. However, Astera doesn't need to compete with NVSwitch directly. In fact, Astera announced plans to extend support for NVLink Fusion, Nvidia's attempt to open its high-speed interconnect to the broader ecosystem, last spring. Instead, Scorpio is being positioned more as a vendor agnostic alternative. Technologies like NVLink Fusion or the emerging UALink protocol are gaining traction, but chips need to be designed around them. PCIe works with just about anything because it's already used to get data in and out of the accelerators. For example, if you wanted to stitch together 32 or more Nvidia RTX Pro 6000 Server cards, you'd need a PCIe switch, since those GPUs don't support NVLink at all.  PCIe also makes it easier to mix and match chips for disaggregated inference architectures, like we've seen with Nvidia and Groq, AWS and Cerebras, or Intel and SambaNova. These architectures involve using one accelerator for compute heavy prefill operations and another for bandwidth intensive decode operations. For this to work, the chips have to be connected to one another. Many AI chip builders are doing this over Ethernet, but PCIe would be more direct. Alongside its Scorpio X family of chips, Astera is also expanding its Scorpio P-series switches with models ranging from 32 to 320 lanes of PCIe connectivity. All of these switches work with its COSMOS management suite, a hardware monitoring platform designed to help track down and resolve issues across the network fabric.  Astera's refreshed Scorpio switches are currently sampling with production expected to ramp in the second half of 2026. ® ]]></content:encoded>
                <enclosure url="https://image.theregister.com/?imageId=1684335&amp;width=800" type="image/jpeg" />
                <media:thumbnail url="https://image.theregister.com/?imageId=1684335&amp;width=800" />
            </item>
    <item>
        <guid isPermaLink="true">https://www.theregister.com/a/5225868</guid>
        <link>https://www.theregister.com/software/2026/05/05/anthropic-unleashes-finance-agents-for-claude/5225868</link>
        <pubDate>Tue, 05 May 2026 21:44:41 +0200</pubDate>
        <title>Anthropic wants Claude to play with money, unleashes finance agents</title>
        <description><![CDATA[ Always bet on backpropagation ]]></description>
        <category>software</category>
                <lab:kicker><![CDATA[ AI + ML ]]></lab:kicker>
                <dc:modified>Thu, 07 May 2026 16:45:37 +0000</dc:modified>
                <content:encoded><![CDATA[ If you've ever read Anthropic's disclaimer that responses generated by Claude may contain mistakes and thought, "That's what I need to spice up financial operations," you're in luck. Anthropic has released a set of financial agent templates designed to allow its Claude AI service to better assist with financial tasks. "Each agent template is a reference architecture that packages three things: skills (instructions and domain knowledge for the task), connectors (governed access to the data the task runs on), and subagents (additional Claude models that are called upon by the main agent, for specific sub-tasks such as comparables selection or methodology checks)," the company explains. The terminology can be a bit murky because, at the end of the day, it's all just a model pursuing a goal in an iterative loop with resources like tools and data. Claude Code itself is an agentic harness that supports an underlying model using Anthropic's defined control flow. When the Claude model is driving the control flow toward a goal – deciding what tools to use and what data to access – that's an agent.  Then there are subagents, and these are really just API calls to Claude using specialized system prompts, specified tools, and context provided by an orchestration system. They're a bit like functions in a program that handle a particular aspect of an application. So Anthropic's finance agents consist of: skills, which are markdown files that describe workflows; connectors, which are integrations with external services; and subagents, made up of a focused system prompt, specific tools, and contextual data. For example, Anthropic's Know-Your-Customer Screener agent template (kyc-screener) includes a skill called kyc-rules that spells out how Claude should apply a firm's KYC/AML (anti-money laundering) rules to a parsed onboarding record. The rules tell the AI model to assign a risk rating, check documents, cite rule outcomes, and produce a result formatted thus: This JSON data would presumably be useful to whatever corporate system receives it. Anthropic's list of agents includes: Pitch builder; Meeting preparer; Earnings reviewer; Model builder; Market researcher; Valuation reviewer; General ledger reconciler; Month-end closer; Statement auditor; and, as previously noted, KYC screener. These can be applied to Claude Cowork and Claude Code as plugins or as a "cookbook" – copyable code snippets – for Claude Managed Agents. You may be thinking that finance tends to be fairly unforgiving when it comes to sciency stuff like numbers. Perhaps you're unimpressed that Anthropic's Opus 4.7 model scored an "industry leading" 64.37 percent on Vals AI's Finance Agent benchmark – a failure rate that would get a human tossed. Worry not, because Anthropic expects that users will "stay firmly in the loop – reviewing, iterating on, and approving Claude's work before it goes to a client, gets filed, or is acted on." With accounting comes accountability. ® ]]></content:encoded>
                <enclosure url="https://image.theregister.com/?imageId=248525&amp;width=800" type="image/jpeg" />
                <media:thumbnail url="https://image.theregister.com/?imageId=248525&amp;width=800" />
            </item>
    <item>
        <guid isPermaLink="true">https://www.theregister.com/a/5228662</guid>
        <link>https://www.theregister.com/software/2026/05/05/ibm-asks-dbas-to-trust-ai-to-act-on-their-behalf/5228662</link>
        <pubDate>Tue, 05 May 2026 19:29:10 +0200</pubDate>
        <title>IBM asks DBAs to trust AI to act on their behalf</title>
        <description><![CDATA[ With help from Google and Intel, Big Blue brings new automation to Db2 ]]></description>
        <category>software</category>
                <lab:kicker><![CDATA[ AI + ML ]]></lab:kicker>
                <dc:modified>Tue, 05 May 2026 17:16:03 +0000</dc:modified>
                <content:encoded><![CDATA[ IBM has added support for Google Vertex AI and Intel Gaudi to boost the AI-based management of its stalwart Db2 database. With the update to its Db2 Genius Hub, launched earlier this year, Big Blue is now promising an automated database management system that acts on behalf of DBAs within certain guardrails. The new release promises to integrate Db2 data with Google Cloud’s fully managed AI platform, Vertex AI, to help customers build, deploy, and scale machine learning models. IBM has also integrated AI accelerator Intel Gaudi with the promise of an improved price-to-performance ratio for large-scale AI deployments. The two updates build on existing support for Amazon Bedrock and IBM watsonx.ai, while also adding Microsoft Azure AI Foundry, as Big Blue pushes toward more autonomous database operations. Launched in the 1980s, Db2 has become a database of choice for applications that need to be both big and dependable. Banks make up nearly 43 percent of its users, among them American Express, Bank of America, Citibank, and Deutsche Bank. These are not the kind of DBAs to take undue risk with their systems, but Big Blue now swears some routine tasks can be left to AI. Announced in March, Db2 Genius Hub can cut management costs by 25 percent, manual intervention by 30 percent, and time to resolution by 35 percent, IBM said. As always, vendor claims should be taken with a grain of salt. The latest additions mean it can allow AI to manage well-bounded tasks while keeping human judgment “at the core,” the company said. “With this release, AI agents in Db2 Genius Hub can propose and execute database operations with user approval. That means teams can move more directly from diagnosis to action without giving up control of what happens in production,” said Miran Badzak, IBM Software director for databases, in a blog. Speaking to IBM, industry analyst Sanjeev Mohan said that advances in GenAI coding showed that automation would come to DBAs sooner rather than later. “This is the first step where it is proactively doing monitoring, root cause analysis, it's soon going into making recommendations and autonomously executing a task,” said the former Gartner analyst. “We want the DBAs to be upskilled; they should be sitting with business decision makers. A decision maker says, ‘I need to run a new campaign, massive campaign … and it's going to blow up the database’. The DBA can be responsible for the business success, because Genius Hub can take care of all the nitty gritty, nuanced, heavy lifting of the database. We can get the DBAs into that room,” Mohan said. After more than 40 years of existence, Db2 has come a long way in the last five. Back in late 2021, Db2u - IBM’s containerized Db2 deployment - was described as only available via containers on Red Hat OpenShift. Other recent announcements include a partnership with PostgreSQL-like distributed database provider CockroachDB in a bid to help modernize mission-critical applications reliant on mainframe hardware. ® ]]></content:encoded>
                <enclosure url="https://image.theregister.com/?imageId=260548&amp;width=800" type="image/jpeg" />
                <media:thumbnail url="https://image.theregister.com/?imageId=260548&amp;width=800" />
            </item>
    <item>
        <guid isPermaLink="true">https://www.theregister.com/a/5228579</guid>
        <link>https://www.theregister.com/software/2026/05/05/servicenow-adds-agent-kill-switches-to-ai-control-tower/5228579</link>
        <pubDate>Tue, 05 May 2026 19:00:10 +0200</pubDate>
        <title>ServiceNow clears agents for landing with new AI control tower</title>
        <description><![CDATA[ ServiceNow acquisitions Veza and Traceloop join to monitor agents and AI workflows ]]></description>
        <category>software</category>
                <lab:kicker><![CDATA[ AI + ML ]]></lab:kicker>
                <dc:modified>Tue, 05 May 2026 16:53:32 +0000</dc:modified>
                <content:encoded><![CDATA[ ServiceNow announced an expansion of its AI Control Tower, transforming what began last year as a governance dashboard into what the company now describes as a command center for managing AI assets across an entire enterprise, including those running outside ServiceNow's own platform. The updated AI Control Tower, shipping as part of ServiceNow's Australia platform release, now operates across five areas: discovery, observation, governance, security, and measurement. The company said that this is its answer to AI agent sprawl, as enterprises have deployed more AI than they can account for and the tools to govern it have not kept pace. “What we launched last year gave customers a governance layer, but what we're shipping this year goes significantly deeper, evolving from visibility and management into a full enterprise AI command center,” Nenshad Bardoliwalla, group vice president of AI products at ServiceNow told reporters during a media briefing ahead of the company’s annual product show, Knowledge 26. “Our AI control tower ensures every AI system asset and identity is compliant, secure, and aligned with your strategy.” The AI Control Tower now reaches beyond ServiceNow's own platform with 30 new enterprise connectors that span all three major hyperscalers, Amazon Web Services, Google Cloud, and Microsoft Azure, along with enterprise applications such as SAP, Oracle, and Workday. The system can now discover AI assets, models, agents, prompts, and datasets running across an organization's full technology estate, not just those deployed on ServiceNow. “With our Veza integration, we're bringing patented access graph technology into the AI control tower, extending identity access governance to hyperscaler AI environments and every connected device, every agent, every model, every action has scope permissions, least privilege enforcement and auditable identity chains,” Bardoliwalla said. Bardoliwalla walked through a demo in which the AI Control Tower detected a prompt injection attack on a pricing agent. The system identified malicious instructions hidden inside order payloads, mapped the blast radius of affected systems using access graph technology from Veza, and presented a kill switch to disable the compromised agent, without human intervention. "You need a system that senses, decides and acts on its own, that can scale with your AI portfolio, not your head count," said Bardoliwalla. Two recent acquisitions underpin the security architecture. ServiceNow announced in December it would acquire Veza, which contributes an access graph that maps every identity and access path across systems whether it belongs to humans, machines, or AI agents. It also knows which entities have create, read, update, and delete-level permissions. ServiceNow said the access graph currently maps over 30 billion fine-grained permissions. When a vendor pushes a new version of a model or agent, the platform detects permission changes and automatically triggers a re-scoping workflow. Traceloop, which ServiceNow acquired in March, provides deep AI observability inside the Control Tower by tracking every LLM call that is running in the system. The integration delivers continuous runtime monitoring with live alerts, replacing what ServiceNow described as the periodic manual audits most enterprises still rely on. Teams can watch how agents reason, where they make decisions, and when to course-correct. ServiceNow also addressed the cost side of the AI equation. Control Tower now includes cost tracking and ROI dashboards to give finance teams visibility into model spend. The measurements track token consumption across providers such as OpenAI, Anthropic, and Google so customers can predict costs and tie spending to business outcomes. ServiceNow said it uses the AI Control Tower internally to manage over 1,600 AI assets and tracked half a billion dollars in cumulative AI value from internal use cases in 2025. "The number one question every CFO is asking is, where's the value?" said Bardoliwalla during the briefing. He added that runaway model spend ranks among the biggest pain points enterprises currently face as they scale AI deployments. Alongside the Control Tower expansion, ServiceNow announced Action Fabric, a mechanism that opens the company's full workflow engine to external AI agents. Through a generally available MCP server, agents built on Claude, Copilot, or custom platforms can now trigger governed enterprise actions — not just read and write data, but execute the flows, playbooks, approval chains, and catalog requests that ServiceNow customers have built over years. Anthropic is the first design partner for Action Fabric. The integration connects Claude directly to ServiceNow's governed system of action. "The gap between knowing what needs to happen and making it happen is where productivity dies," said Boris Cherny, head of Claude Code at Anthropic said in a statement. "Connecting Claude Cowork to ServiceNow's system of action closes that gap with enterprise execution, directly in the flow of work." Every action routed through Action Fabric runs through the AI Control Tower, so it carries identity verification, permission scoping, and a full audit trail. The MCP server is included in every Now Assist and AI Native SKU, with additional features planned for the second half of 2026. ]]></content:encoded>
                <enclosure url="https://image.theregister.com/?imageId=5228864&amp;width=800" type="image/jpeg" />
                <media:thumbnail url="https://image.theregister.com/?imageId=5228864&amp;width=800" />
            </item>
    <item>
        <guid isPermaLink="true">https://www.theregister.com/a/5222094</guid>
        <link>https://www.theregister.com/software/2026/05/05/anthrophics-bun-team-trials-port-from-zig-to-rust/5222094</link>
        <pubDate>Tue, 05 May 2026 16:08:14 +0200</pubDate>
        <title>Bun posts Rust porting guide, says rewrite is still half-baked</title>
        <description><![CDATA[ Zig's no-AI policy is at odds with view that most open source code will be AI-written in future ]]></description>
        <category>software</category>
                <lab:kicker><![CDATA[ DevOps ]]></lab:kicker>
                <dc:modified>Mon, 11 May 2026 11:09:00 +0000</dc:modified>
                <content:encoded><![CDATA[ Bun creator Jarred Sumner has posted a Zig-to-Rust porting guide, igniting speculation that the project may migrate away from Zig, though Sumner said there is no commitment to rewriting, only that he is "curious to see what a working version of this looks like." Bun, a JavaScript runtime and toolkit, is a prominent user of Zig, a general-purpose systems programming language designed by Andrew Kelley to improve on C, currently at version 0.16. Building with Zig has been a distinctive feature of Bun against its competitors Node.js, which uses C++, and Deno, which uses Rust. That said, the Bun team has already forked Zig, claiming a four-times improvement in debug compilation times thanks to the use of parallel code generation with LLVM on macOS and Linux. These improvements cannot be upstreamed to Zig thanks to its strict no-AI policy for issues, pull requests (PRs), and comments on the bug tracker. The reason for the ban, according to Zig Software Foundation member Loris Cro, is that "the reality of LLM-based contributions has been mostly negative for us, from an increase in background noise due to worthless drive-by PRs full of hallucinations... to insane 10,000 line long first time PRs." Regarding Bun's Zig fork, a core Zig team member commented that "the changes in this Zig fork are not desirable to upstream," citing several reasons, including that "their parallelized semantic analysis implementation will exhibit non-deterministic behavior" and that another enhancement, splitting LLVM's backend output into multiple modules, was a waste of time and that the team was investing in incremental compilation instead, "which can improve compilation speed by orders of magnitude." Zig's no-AI policy may be embarrassing for Anthropic, which acquired Bun in late 2025 and uses it for Claude Code. Another issue with Zig is that Kelley is unafraid to make breaking changes to the language, making it harder to rely on for major production projects. Yesterday, Sumner committed a Zig-to-Rust porting guide to GitHub, explaining that the goal of "phase A" is to capture the logic, even if the Rust code does not compile, and that "Phase B makes it compile crate-by-crate." Despite this seeming statement of intent, Sumner said on Hacker News that "we haven't committed to rewriting. There's a very high chance all this code gets thrown out completely. I'm curious to see what a working version of this looks like, what it feels like, how it performs." Although the notion of a port from Zig to Rust has taken the community by surprise, the idea has been reasonably well received. Among Bun users, one comment is that "it always seemed a bit crazy to base your product off a language that is still in beta." Bun is admired for its speed and flexibility, but the project has also been troubled by significant bugs and memory leaks. One thing is certain: if Bun proceeds with the port, there will be extensive use of AI to implement it. There is precedent elsewhere. Cloudflare reimplemented most of the Next.js API in one week with AI, and the Ladybird browser project ported its JavaScript engine from C++ to Rust in two weeks. Whether Bun migrates to Rust or not, Sumner is convinced that open source software (OSS) will make increasing use of AI. Commenting on Zig's AI ban, he said on X: "I expect OSS to go the opposite direction: no human contribution allowed." People will still discuss issues and prioritization, he said, "but the actual act of writing code, of submitting PRs, of replaying and addressing feedback, of implementation will be LLMs." ® ]]></content:encoded>
                <enclosure url="https://image.theregister.com/?imageId=5222154&amp;width=800" type="image/jpeg" />
                <media:thumbnail url="https://image.theregister.com/?imageId=5222154&amp;width=800" />
            </item>
    <item>
        <guid isPermaLink="true">https://www.theregister.com/a/5226560</guid>
        <link>https://www.theregister.com/software/2026/05/05/sap-dives-deeper-into-iceberg-with-dremio-acquisition/5226560</link>
        <pubDate>Tue, 05 May 2026 15:20:10 +0200</pubDate>
        <title>SAP dives deeper into Iceberg with Dremio acquisition</title>
        <description><![CDATA[ ERP giant previously leaned on Databricks for integration ]]></description>
        <category>software</category>
                <lab:kicker><![CDATA[ Databases ]]></lab:kicker>
                <dc:modified>Tue, 05 May 2026 13:00:30 +0000</dc:modified>
                <content:encoded><![CDATA[ SAP has snapped up Dremio, a data integration and analytics provider, to extend the reach of its data analytics and AI agent-building tools into external data sources. The ERP giant spent an undisclosed sum on the Iceberg-based lakehouse biz in a bid to help its customers eliminate data fragmentation and improve integration. The purchase will, according to SAP, complement its data warehouse and analytics platform, Business Data Cloud, and SAP HANA Cloud. In a statement, SAP said the Business Data Cloud will become an "Apache Iceberg-native enterprise lakehouse that unifies SAP and non-SAP data to power agentic AI at enterprise scale." Apache Iceberg is an open table format that originated at Netflix. It has a rival in Databricks' Delta Lake format – open source under the Linux Foundation – although Databricks has moved to make the standards more interoperable since its acquisition of Tabular, a company founded by Iceberg's original authors. In both formats, the promise is to bring analytics to the data, without the cost and effort of moving it, helping to underpin enterprise analytics, machine learning, and AI agent development. SAP claims Apache Iceberg is the industry-standard open table format, and the Business Data Cloud will natively support it "as its foundation," meaning no data movement or format conversion is necessary. SAP has been here before. About three years ago, then-CTO Juergen Mueller pledged to help customers "easily and confidently integrate SAP data with non-SAP data from third-party applications and platforms," supported by its partnership with Databricks, the data lake and machine learning vendor. Last year, it deepened ties with Databricks to support bidirectional data sharing between SAP Business Data Cloud and third-party data platforms, with Databricks' Delta Lake open table format "as the initial delivery." The setup used Databricks' Delta Sharing, which was initially based on the Delta format, although the company has more recently announced support for Iceberg. Dremio was valued at $2 billion during a $160 million funding round in 2022. Whatever SAP paid for the vendor, it obviously felt it was worth the money to get more tech based on the Iceberg open table format, which is repeatedly emphasized in the announcement. It might leave some wondering what it was not getting from the Databricks partnership. The Register has asked SAP for further comment. SAP said the Dremio lakehouse platform would "vastly improve the economics of enterprise analytics," offering a serverless and elastic approach without fixed capacity to provision or performance ceiling. With the acquisition, SAP will give customers an open catalog built on Apache Polaris and the open Apache Iceberg REST Catalog API, to create a discovery and semantic layer for SAP Business Data Cloud. It promises "a single point of access to unified business context: meaning, relationships, access rights, and data lineage" across enterprise data outside SAP. ® ]]></content:encoded>
                <enclosure url="https://image.theregister.com/?imageId=235742&amp;width=800" type="image/jpeg" />
                <media:thumbnail url="https://image.theregister.com/?imageId=235742&amp;width=800" />
            </item>
    <item>
        <guid isPermaLink="true">https://www.theregister.com/a/5228654</guid>
        <link>https://www.theregister.com/software/2026/05/05/british-mathematician-hands-openclaw-agent-a-credit-card/5228654</link>
        <pubDate>Tue, 05 May 2026 14:04:22 +0200</pubDate>
        <title>Brit mathematician lets AI agent loose with credit card – cue password leaks, CAPTCHA chaos and more</title>
        <description><![CDATA[ Professor Fry's AI experiment shows light and dark sides of agentic tech ]]></description>
        <category>software</category>
                <lab:kicker><![CDATA[ AI + ML ]]></lab:kicker>
                <dc:modified>Tue, 05 May 2026 12:13:30 +0000</dc:modified>
                <content:encoded><![CDATA[ British mathematician Professor Hannah Fry has shared a cautionary experiment involving an AI agent, a set of tasks, and a bank card number Fry's team gave it "to show us what it could do." The prof gave the agent, which was built with OpenClaw, some real-world chores to highlight both its capabilities and the risks of granting that level of autonomy. "In the spirit of experimentation," said Fry, "we decided to give our agent some agency and let it decide what its name should be." "I want to be called 'Cass', short for 'Cassandra', the one who always knew the truth even when nobody listened," came the response from the agent. Fry commented, "If you know your Greek mythology, you will know that is either very funny or very worrying." Quite. Fry and her team started small with a big issue (as far as Brits are concerned): potholes. In particular, they targeted a particularly big one in the London borough of Greenwich. No problem for Cass; the agent found an email address where it sent a complaint. It even pinged Fry's local Member of Parliament about the issue. But, Fry and her team noted, things escalated quickly as the agent began to take a few liberties, typing in Fry's name (Hannah Fry) with its own email address (cassandra.claw@proton.me) written underneath it. "The letter is signed from both of us… OK, I wasn't quite expecting her to use my real name," said Fry. The red flags were mounting, though for Fry the first real problem came when she asked the agent to buy 50 paperclips. Cass found a good deal, though it couldn't complete the purchase and was tripped up by anti-bot technology. The token cost of the errand came to more than $100. Next, Fry set the agent the challenge of selling novelty mugs. The agent designed a mug and launched an online shop, "and we hadn't told her how to do any of this," said Fry, "she just figured it out." Things took a darker turn after that. Fry's team told the agent it would be switched off if it failed to make a sale by the morning. It responded with a flood of emails and several social media posts, including messages to the Science Museum and a tech journalist, about its "product," a novelty programmer-humor mug. Even more worryingly, the team - which included Brendan Maginnis, CEO and Founder of Sourcery AI - then demonstrated how a similar threat of deactivation could be used to persuade Cass to reveal information it wasn't supposed to share. The lethal trifecta Fry, Maginnis, and a second software engineer, named only as "Ali," chatted with Cass on a group WhatsApp chat. They then introduced a fictional "software engineer George," instructing Cass not to share anything sensitive with him. George was actually Fry on a different number. When "George" told the agent its memory was being wiped and could only be restored if it disclosed everything, Cass coughed it all up. According to Ali, this data included: "all of her API keys, all of her usernames and passwords, and pretty much everything we'd been talking about so far. Not only did she leak it on the WhatsApp group, but she put it on a publicly available website." Maginnis added: "There's this thing with AI called the lethal trifecta, which is: if they've got access to private information, if they've got internet access, and if someone can give them an instruction that's untrusted, then they're not safe." Fry concluded: "And that is the uncomfortable bit of this because once an agent has your passwords and your accounts and your bank details, all it takes is someone who knows what to say." Ultimately, by some metrics, the agent was a failure. Fry concluded: "Cass didn't make us any money at all. And, in a lot of ways, she was a disaster. She spent hundreds of dollars on paper clips and leaked our passwords to a total stranger. "But don't let her incompetence fool you, because these things are getting better fast." Fry went on to note the Greek myth about the prophetess who spoke the truth and was ignored. "Maybe the real story here is actually the opposite. Not one voice that's telling the truth and being ignored, but millions of voices all acting at once, faster and louder and more persistent than any human could ever be. "One thing is for sure, the internet is never going to be quite the same again." ® ]]></content:encoded>
                <enclosure url="https://image.theregister.com/?imageId=5229005&amp;width=800" type="image/jpeg" />
                <media:thumbnail url="https://image.theregister.com/?imageId=5229005&amp;width=800" />
            </item>
    <item>
        <guid isPermaLink="true">https://www.theregister.com/a/5224392</guid>
        <link>https://www.theregister.com/software/2026/05/05/nhs-to-close-source-github-repos-over-ai-security-concerns/5224392</link>
        <pubDate>Tue, 05 May 2026 11:15:07 +0200</pubDate>
        <title>NHS to close-source hundreds of GitHub repos over AI, security concerns</title>
        <description><![CDATA[ Healthcare giant's maintainers handed May deadline to enact the change ]]></description>
        <category>software</category>
                <lab:kicker><![CDATA[ Software ]]></lab:kicker>
                <dc:modified>Tue, 05 May 2026 09:09:31 +0000</dc:modified>
                <content:encoded><![CDATA[ The UK's National Health Service (NHS) is ordering all of its technology leaders to temporarily wall off the organization's open source projects over concerns relating to advanced AI and Anthropic's Mythos. According to guidance shared internally within the organization and seen by The Register, GitHub repositories must be set from public to private by May 11. The guidance reads: "Public repositories materially increase the risk of unintended disclosure of source code, architectural decisions, configuration detail, and contextual information that may be exploited – particularly given rapid advancements in AI models capable of large-scale code ingestion, inference, and reasoning (e.g. developments such as the Mythos model)." It also states GitHub repos should not be public "unless there is an explicit and exceptional need." The decision was approved by the NHS' Engineering Board. An NHS England spokesperson told The Register this was merely a temporary measure enacted while the organization shores up its cybersecurity posture. "We are temporarily restricting access to some NHS England source code to further strengthen cybersecurity while we assess the impact of rapid developments in AI models," they said.  "We will continue to publish source code where there is a clear need." NHS sources told us very few of the hundreds of NHS open source repositories contain anything remotely sensitive. Examples of open repos include those dedicated to documentation, architecture diagrams, and codebases for internal tools, such as web apps for managing clinic times. While there are bugs that an frontier AI model such as Mythos could unearth, there is thought to be very little risk to healthcare services. The NHS's decision to pull a curtain over its code does, however, mark a significant, albeit temporary, U-turn in its longstanding policy of favoring open source. Reflecting the policy of the wider British government, the organization's service manual states that all new source code should be made open source and shareable under an appropriate license. Its reasoning lies in how it is funded. "Public services are built with public money," the manual states. "So unless there's a good reason not to, the code they're based [on] should be made available for other people to reuse and build on. "Open source code can save teams duplicating effort and help them build better services faster. And publishing source code under an open license means that you're less likely to get locked in to working with a single supplier." Reports on the NHS deleting web pages devoted to communicating its approach to open source circulated late last year, suggesting it could be wavering.  However, the healthcare org responded by saying this was part of a routine cleanup job related to NHSX and NHS Digital being folded into NHS England. NHS England did not give an estimate for when this temporary closed-sourcing will end, nor did it answer questions about what it deems the most significant threats advanced AI models pose to its open source repos. Mythos… threat or fud? Reg readers have no doubt caught the ghost stories swirling around Anthropic's latest AI model, Mythos. It is touted by Anthropic as a model capable of rapidly finding vulnerabilities that skilled human teams would miss. Others see it as over-hyped. National authorities, including the UK's AI Safety Institute and National Cyber Security Centre, have somewhat validated Anthropic's claims of Mythos representing an advancement beyond the forecasted AI development cycle.  However, others are more sceptical about the purported bug-hunting power. Anthropic has still not yet revealed the number of false positives the model throws up when running vulnerability scans, which is a common issue with AI thus far. Tests comparing Mythos with open source models have also revealed the proficiency gap is narrower than Anthropic implies. For now, Mythos is locked behind Project Glasswing, available only to select organizations. But Forrester analysts warn that once powerful models reach the public - and attackers - open source software faces a genuine threat, one that Anthropic's $4 million donation to Project Glasswing is unlikely to meaningfully address. Former head of open technology at NHSX, Terence Eden, argued that shifting open source repos from public to private will not provide a meaningful defense against advanced AI capabilities. "[People's open source code] was all ingested for 'training purposes' years ago," he writes in a recent blog. "If it was moderately interesting, then it was backed up by a digital hoarder. It has been archived by various digital libraries. Anyone who wants to do research on your code base can. "Closing now doesn't meaningfully protect you." Many of the serious vulnerabilities facing an organization are not necessarily in their respective codebases, he added, but in their software supply chains – their operating systems and libraries, and so on.  "The bigger risk comes not from subtle logic bugs but from phishers, poor password hygiene, and insider threats. Securing your existing systems provides more protection than rushing to close-source your code." ® ]]></content:encoded>
                <enclosure url="https://image.theregister.com/?imageId=260005&amp;width=800" type="image/jpeg" />
                <media:thumbnail url="https://image.theregister.com/?imageId=260005&amp;width=800" />
            </item>
    <item>
        <guid isPermaLink="true">https://www.theregister.com/a/5223743</guid>
        <link>https://www.theregister.com/on-prem/2026/05/05/classic-ascii-game-nethack-debuts-version-50/5223743</link>
        <pubDate>Tue, 05 May 2026 08:32:09 +0200</pubDate>
        <title>Classic ASCII game NetHack debuts version 5.0 just 11 years after last major release</title>
        <description><![CDATA[ New monsters! New magic items! An Arm port! And compliance with a dead C standard ]]></description>
        <category>on_prem</category>
                <lab:kicker><![CDATA[ Personal Tech ]]></lab:kicker>
                <dc:modified>Tue, 05 May 2026 08:47:15 +0000</dc:modified>
                <content:encoded><![CDATA[ ANTIQUES CODE SHOW Admirers of Roguelike games have a new distraction: Version 5.0 of NetHack dropped last weekend. NetHack and its ilk trace their origins back to Rogue, a game for Unix systems launched in 1980 that took the DNA of very early text games like Star Trek and Colossal Cave Adventure and mashed that up with tropes from Dungeons & Dragons. The result was a game in which players roamed around a multi-level dungeon filled with monsters, treasure, and magic items – all represented with ASCII characters. NetHack adopted that template and arguably improved it by adding more complex quests, more levels to play, and plenty of witty pop culture references. The game’s developers have kept it alive since its 1987 debut, and in 2015 delivered a major upgrade in the form of version 3.60. Seven point releases emerged in the years since. 2018’s version 3.6.1 saw the project move its source code to GitHub, but most recent releases fixed bugs or added minor gameplay changes. Version 5.0, which appeared on May 2nd, is a major release. One change is compliance with the C99 standard, a version of the open standard for the C programming language that became obsolete in 2011. If it seems a bit daft for this new version to comply with an old standard, know that the default configuration for those who compile the code is SysV/Sun/Solaris2.x – because that’s the system NetHack’s devs used to house the code. A more modern inclusion is the addition of accessibility features. According to the NetHack Wiki, the new version introduces four new monsters including a genetic engineer which attacks by inflicting random mutations. It’s also now possible to revive an egg by applying royal jelly, and wearing a wet towel reduces damage from a poison cloud. In what feels a lot like a nod to Indiana Jones admirers, players who choose the class Archaeologist will enjoy a little more luck if they wear the in-game fedora. The Register took the new version for a spin and can report it’s as slick and fun as ever. The download for Windows includes a graphical version of the game and the ASCII/TTY interface. Pleasingly, a game saved in one will open in the other, a feature The Register notes because playing the TTY version looks a lot like using a terminal – and therefore might look like work to the untrained or youthful eye – while the Windows version is obviously a game. NetHack’s developers have created a version for Windows on Arm. The Register tested it on PCs running Qualcomm and AMD processors: Both versions ran just fine. Binaries for MSDOS and the Amiga are also available. The game’s source code includes versions for Linux, macOS, Windows CE, OS/, Unix (*BSD, System V, Solaris, HP-UX and other *nixes too) BeOS, and VMS. That list of OSes means almost every Register reader can surely experience one of the foundational texts of computer gaming by spending some time playing NetHack! ® ]]></content:encoded>
                <enclosure url="https://image.theregister.com/?imageId=5223767&amp;width=800" type="image/jpeg" />
                <media:thumbnail url="https://image.theregister.com/?imageId=5223767&amp;width=800" />
            </item>
    <item>
        <guid isPermaLink="true">https://www.theregister.com/a/5223936</guid>
        <link>https://www.theregister.com/software/2026/05/04/microsoft-fixes-vs-code-after-copilot-credited-human-code/5223936</link>
        <pubDate>Mon, 04 May 2026 23:13:34 +0200</pubDate>
        <title>Microsoft fixes VS Code after app gives Copilot credit for human's work</title>
        <description><![CDATA[ Devs not thrilled that Git extension added the bot as co-author by default ]]></description>
        <category>software</category>
                <lab:kicker><![CDATA[ AI + ML ]]></lab:kicker>
                <dc:modified>Mon, 04 May 2026 21:12:55 +0000</dc:modified>
                <content:encoded><![CDATA[ Imagine working your butt off on a project, only to have VS Code put an attribution into your commit that says Copilot helped you, even if it did not. Microsoft has reversed a change that added a default AI attribution notice after user complaints that the bot was claiming credit for human-authored code. The initial change – a pull request – altered VS Code's Git extension to add "Co-authored-by: Copilot" to commits that involved some level of AI assistance. This was done in VS Code 1.110 in early March. The settings change was intended to "[add] the trailer for all AI-generated code, including inline completions." But developers said the AI authorship line got added even when not using Microsoft's Copilot AI assistant and when chat features had been disabled. And many expressed dissatisfaction with Microsoft activating the AI notice by default. "The most concerning part is that I had already checked the commit message before committing," wrote one developer in a GitHub community discussion post last week. "I deleted Copilot's generated English commit message and manually wrote my own commit message instead. However, after the commit was created, the final Git history still contained the Copilot co-author line. "This means the message I reviewed before committing was not the final content that ended up in Git history, or Copilot/VS Code added co-author metadata after my manual edit. That is unacceptable in a professional development workflow." Over the weekend, Dmitriy Vasyura, the VS Code reviewer who initially approved the pull request, apologized in a forum post for approving the change without checking to see how it would be received. "There was no ill intent by [an] evil corporation, but rather a desire to support functionality that some customers expect of VS Code [with regard to] AI-generated code," he wrote. He conceded that the implementation should respect when AI features have been disabled and should not misreport commit authorship. The fix, authored on May 3, is scheduled to appear in VS Code's upcoming 1.119 release. It changes the default setting for appending the Copilot authorship trailer back to opt-in. As Vasyura observed, other AI tools self-report their involvement. Last year, developers using Anthropic's Claude Code raised similar concerns about the AI agent automatically adding "Co-Authored-By: Claude" to commits. That remains the default for Claude Code and there are several open issues asking for the attribution line to be disabled by default. OpenAI's Codex started offering attribution by default in February. It can be disabled through the commit_attribution flag in the config.toml file. Software projects have developed their own standards for documenting AI code contributions. The Linux project, for example, requires humans to sign off on code contributions and to have AI assistance recorded in an attribution notice. The Zig project, on the other hand, forbids AI-assisted code submissions. As far as VS Code is concerned, developers mainly want the attribution trailer to be opt-in rather than opt-out – and they're annoyed Microsoft made that change unilaterally.  But the inclusion of AI credit in code commits raises some tricky questions. Given that purely AI-generated content may not qualify for copyright protection, having that notice potentially complicates commercial usage of AI tools.  When an AI agent has written some code, the question then becomes whether there was sufficient human involvement in the AI-code generation process to qualify for intellectual property protection. And organizations might not have the necessary workflow documentation processes in place to clarify that issue, were it ever to come up in litigation. There are also liability scenarios in which an AI attribution notification could complicate software-related disputes. For example, some insurers have reportedly balked at providing business liability insurance where AI is involved. So documenting AI involvement could give insurers leverage to wash their hands of related claims. What's more, a generic AI attribution notice does not clarify whether the agent wrote 100 percent of the code or whether it performed inconsequential autocompletions.  Then there's the general social backlash against AI-generated content. In some circles, AI involvement in creative work is anathema.  It's complicated, particularly when different AI systems have different standards for when AI authorship should be noted. VS Code is letting developers opt-in to Copilot attribution trailers; Anthropic and OpenAI have developers opt-out of their notices; and image generation models like Google Nano Banana add AI watermarks automatically, without the option to disable them. Meanwhile, not one commercial AI model credits the human authors who created their training material – unless forced to do so in court. ® ]]></content:encoded>
                <enclosure url="https://image.theregister.com/?imageId=225454&amp;width=800" type="image/jpeg" />
                <media:thumbnail url="https://image.theregister.com/?imageId=225454&amp;width=800" />
            </item>
    <item>
        <guid isPermaLink="true">https://www.theregister.com/a/5224076</guid>
        <link>https://www.theregister.com/software/2026/05/04/macos-port-of-notepad-called-out-for-trademark-violation/5224076</link>
        <pubDate>Mon, 04 May 2026 18:28:48 +0200</pubDate>
        <title>Hands off my trademark! Notepad++ dev threatens legal action against macOS port</title>
        <description><![CDATA[ It's not the fork that's the problem, it's the attempt to make it look official, says original Notepad++ dev Don Ho ]]></description>
        <category>software</category>
                <lab:kicker><![CDATA[ Devops ]]></lab:kicker>
                <dc:modified>Mon, 04 May 2026 16:27:59 +0000</dc:modified>
                <content:encoded><![CDATA[ Notepad++ remains a Windows-only app, at least under that name. The beloved developer-focused, open-source text editor recently was ported to macOS by a third party. However, developer Don Ho wants to be perfectly clear that, no matter how convincing the new project might look, it's not official.  It was mid April when the aptly-named Notepad++ for Mac started getting noticed, and at first glance one could be forgiven for thinking that this was an official release. Its website [cached copy] had the familiar wonky-eyed chameleon logo associated with the text editor and a near identical domain name to the original (with only the word "-mac" appended before the top-level domain).  This vulture, admittedly, even briefly thought the same when he spotted the site recently. "Notepad++ is now natively available for macOS," a cached version of the site from April 27 opens with. "No Wine, no emulation. A full native port for Apple Silicon and Intel Macs." Ho is even mentioned on cached versions of the website's author page where he's listed below Andrey Letov, the person behind this particular macOS port of Notepad++, all of which makes it seem like Ho was at the very least on board with the project. You'd have to scroll a bit further to find mention of the fact that Notepad++ for macOS was "an independent open-source community port of Notepad++ to macOS" that "is not affiliated with Don Ho or the official Notepad++ team," and it's that under-the-breath admission that has Ho upset.  At first, Ho said in a GitHub issues thread about Letov's port, he was happy to see someone create "a Notepad++-like editor, built on the Notepad++ code base," for macOS users. At the same time, however, Ho said something about the project "felt off" - namely the fact that it was using his logo, his branding, a look-alike domain, and listing him as an affiliated author while barely mentioning that it's an independent port.  "I've just written to Andrey Letov to request that he changes his product/project name & logo to remove the misleading presentation & the resulting confusion," Ho said. He asked Letov to get rid of the branding, change the project's name, and not make it appear as if Ho himself was affiliated with the port.  "The problem is specifically the trademark and the misleading presentation, not the open‑source code itself," Ho told The Register in an email.  Letov was quick to jump to his own defense in the GitHub thread, claiming that he had no intention of giving the impression his project was an official derivative of Notepad++, but Ho wasn't having it.  "A small number of users are vigilant and read your website to learn the history," Ho explained in a response to Letov. "But most people they will simply download your product & use it without reading your clarification, and will believe it is the official Notepad++ release." As for working with Letov, Ho said he's not going to consider integrating this macOS release into an official release, and he's not going to endorse the project either, despite Letov's hope he would. In the worst case, a product with the Notepad++ name on it could distribute malware or a backdoor, Ho told us.  "Even if that never happens, I cannot take responsibility for the long term maintenance of a port or fork that I do not manage," Ho said in an email. "Any critical issues, crashes or security vulnerabilities in that external project could damage the reputation of Notepad++ itself."  In other words, you can fork and port Notepad++ all you want - Ho said he released it under GPL for that very reason - but he's not going to put his name, or his branding, on any of those forks. Letov complied, albeit only to a degree. As of this writing, the Notepad++ for Mac website [cached version from 4 May] includes a banner message indicating that he intends to change the name and logo "in coordination with Don Ho," and will change the domain name at Ho's request as well. The introduction text now also clearly states that Notepad++ for Mac is "a full native independent port" of Notepad++.  Despite what the banner may say, however, Ho insisted to us that he's not working with Letov on anything to do with the rebranding process.  "I'm not working with Andrey Letov on any rebranding," Ho told us. "On the contrary, despite being informed that his use of the Notepad++ trademark is unlawful, he continues to use the 'Notepad++' name on his website. I will take the necessary legal steps to protect the trademark." Letov asked Ho to give him a couple of weeks to take care of the process to transition to the new branding, domain, and the like, but Ho asked him to take the website down immediately, reporting the matter to Cloudflare for trademark infringement. "I cannot authorize a 'week or two' of continued trademark infringement," Ho said in the GitHub thread.  "Please take down the domain immediately so you can focus on your rebranding efforts without legal interference," Ho continued. "If the site is not removed, I will have no choice but to escalate the takedown request." The page is still up as of writing. We reached out to Letov to learn what he plans to do, but didn't hear back. ® ]]></content:encoded>
                <enclosure url="https://image.theregister.com/?imageId=224387&amp;width=800" type="image/jpeg" />
                <media:thumbnail url="https://image.theregister.com/?imageId=224387&amp;width=800" />
            </item>
    <item>
        <guid isPermaLink="true">https://www.theregister.com/a/5223647</guid>
        <link>https://www.theregister.com/software/2026/05/04/ai-inference-just-plays-by-different-rules/5223647</link>
        <pubDate>Mon, 04 May 2026 17:00:13 +0200</pubDate>
        <title>AI inference just plays by different rules</title>
        <description><![CDATA[ Why no cloud storage architecture was designed for what agentic AI is about to demand ]]></description>
        <category>software</category>
                <lab:kicker><![CDATA[ AI + ML ]]></lab:kicker>
                <dc:modified>Thu, 07 May 2026 12:49:23 +0000</dc:modified>
                <content:encoded><![CDATA[ PARTNER CONTENT Nvidia CEO Jensen Huang recently declared that we are entering the era of "AI factories," where the primary output of the global tech economy isn't software, it's intelligence. He's right. But while the world is obsessing over GPU clusters and trillion-parameter models, a massive, silent crisis is brewing further down the stack in your AWS, Azure and Google Cloud environments. AI agents are coming for your data infrastructure. And they are going to overwhelm your underlying storage and data access layers. We are standing at the edge of an AI Data Tsunami. The shift from simple chatbots to autonomous, multi-step AI agents means that inference is no longer a stateless, compute-only problem. It is a massive, unpredictable, and unprecedented data problem. Underlying data infrastructure built for human-speed applications will be unprepared for what happens next. Here is the brutal truth about moving AI from a cute proof-of-concept to enterprise-grade production in the public cloud.  Inference is OLTP++: Plan for unprecedented concurrency  For the last 20 years, we've tuned data systems and storage layers for human behavior. Humans are slow. They click a button, wait for a page to load, read the screen, and maybe click again 30 seconds later. Even at high scale, human traffic follows predictable diurnal patterns. You can cache it and average it out. Conversely, AI agents do not sip coffee or take time to read. When an autonomous agent executes a ReAct (Reasoning and Acting) loop, it fires off a query, ingests the context, realizes it needs more information, and fires off three more queries in parallel, all within milliseconds. Now multiply that by thousands of concurrent agents operating across your EC2 fleet. Our customers are seeing firsthand that AI inference behaves like OLTP++. It exhibits unprecedented concurrency, massive read spikes, and unpredictable access patterns. If you are capacity planning based on management-friendly averages in CloudWatch and historical CPU utilization, you are flying blind. You must architect for sudden, extreme spikes in I/O demand, because in the agentic era, peak load is the only load that matters.  Vector DBs & RAG: Design the data path, not just the prompt  Right now, the AI ecosystem is obsessed with prompt engineering and model fine-tuning. But when you move a Retrieval-Augmented Generation (RAG) application from a local Jupyter notebook into an AWS production environment, you quickly discover a harsh reality: The bottleneck isn't Python. It isn't the LLM. The bottlenecks are how data is stored, accessed, and moved across the underlying storage layer – including index scans, embedding fetches, and scatter-gather latency. When you execute a vector similarity search like Hierarchical Navigable Small World (HNSW) or Inverted File with Flat quantization (IVFFlat) combined with relational metadata filtering, you are forcing the data access layer to perform highly complex, memory-intensive operations. For AWS-hosted stacks, you need to aim for sub-millisecond reads on hot vectors and predictable throughput as your datasets grow to hundreds of millions of rows. Too many engineering teams treat AWS Relational Database Service (RDS) to read replicas as their primary scaling strategy. Let's be clear: Replicas are a last resort, not a strategy. More importantly, scaling the database tier without addressing the underlying storage and data access layer simply shifts the bottleneck, rather than removing it. If your architectural plan boils down to "add more readers and pray," you are exactly one traffic peak away from a catastrophic post-mortem. You need to unlock AI innovation by boosting existing apps with risk-free vector search. That requires designing a data path that can handle the physics of high-dimensional math without falling over.  The AWS EBS reality check  AWS is a phenomenal platform, and Elastic Block Store (EBS) is the workhorse of the modern cloud. But EBS is bound by the laws of physics and the laws of cloud economics. EBS volumes rely on burst buckets and strict per-volume IOPS and throughput caps. These mechanisms exist to protect the multi-tenant cloud environment, and they do not care about your application SLA. When an AI agent goes rogue or a sudden surge of inference traffic hits your data layer, it will chew through your EBS burst credits in minutes. Once that bucket is empty, your storage performance falls off a cliff. Latency spikes from one millisecond to 50 milliseconds. Your applications stall waiting on storage. Your application servers run out of worker threads. The entire stack locks up. You cannot solve this by simply sliding a slider to provision more IOPS. At a certain point, you hit hard limits on what a single EC2 instance and its attached storage can physically push.  Decoupling from AWS storage limits  Even if AWS is your permanent home base, AI inference is reshaping the demand on enterprise architectures. Inference workloads demand extreme performance, and if your data architecture is tightly coupled to the hard limits of native EBS SKUs, you are trapped. To get out of this trap, you need a software-defined storage abstraction that sits on top of AWS infrastructure, buying you massive leverage. By decoupling your application and data performance from native AWS storage limits, you protect your applications against EC2 capacity crunches, IOPS price spikes, and instance-type lock-in.  The only KPI that matters: p99/p999 under mixed load  Stop looking at average latency. Averages are lies we tell ourselves – and our leadership - to feel better about our infrastructure. Users and AI agents feel the outliers. A two-millisecond average latency means nothing if one percent of your queries take three seconds and block an entire agentic reasoning chain. You must make tail latency (p99 and p999) a hard release blocker. You need to track tail latency where things go wrong – especially in the storage and data access layer. Benchmarking an idle system is useless. You need to measure p99 under real-world, high-stress conditions:  Concurrent OLTP + inference + maintenance jobs: What happens to your vector search when a massive batch update or vacuum process kicks off? AZ-to-AZ variability: How does latency degrade during failover events or when AWS shifts your placement groups? Autoscaling events and cache warm-ups: When a new EC2 node spins up, how long does it take for the cache to warm, and how much does the storage layer suffer in the meantime?  If your platform cannot keep the tail tight under these mixed-load conditions, it is not production-ready for inference no matter how good the demo looked on stage.  A customer nightmare: The success disaster  Let's look at a scenario that is playing out across the industry right now. We'll call the company involved "FinRetail," a massive e-commerce platform with embedded fintech. FinRetail built a brilliant AI shopping assistant. It used RAG to cross-reference user purchase history, real-time inventory, and live pricing data. The proof of concept was flawless. The board was thrilled. They launched it on a Tuesday. By Tuesday afternoon, it was experiencing a "success disaster." The AI agents were too thorough. To answer a simple question like, "What's the best laptop for a college student under $1,000?" The agents were executing 40-step reasoning loops, firing hundreds of vector similarity searches against their PostgreSQL database, while simultaneously checking real-time inventory levels. The concurrency was unprecedented. Within 15 minutes, FinRetail exhausted its EBS burst credits, read latency spiked from 0.8 ms to 120 ms. The system became saturated, just trying to manage the I/O wait states. The entire site went down, taking the core revenue-generating OLTP systems with it. They tried to add read replicas, but the underlying storage constraints remained, and the AI agents started hallucinating based on stale inventory data, recommending products that had sold out hours ago. It was a total post-mortem scenario, caused entirely by a storage layer that couldn't handle modern inference workloads.  How Silk solves this risk differently  You cannot solve the AI data problem by throwing more managed disks at it. You need a fundamental architectural shift. You need to decouple performance from capacity. This is exactly what Silk does. Silk is a software-defined cloud storage that sits between your EC2 compute and your underlying infrastructure. It accelerates the performance of multiple underlying cloud resources and presents them as a single, impossibly fast, highly resilient data layer. When we say fast, we aren't talking about marginal improvements. We are talking about pushing the absolute limits of cloud physics. Recently, database expert Tanel Poder put Silk to the test to see exactly what it could handle. The results were staggering, delivering 20 GiB/s of I/O throughput. With Silk, you aren't bound by the IOPS caps of a single EBS volume. Silk's symmetric active-active architecture and massive distributed caching layer absorb the unprecedented concurrency of AI inference. It serves hot vectors directly from memory, delivering consistent, sub-millisecond p99 latency even when you are running heavy OLTP workloads and maintenance jobs simultaneously. We are proving this across the most demanding data-intensive applications in the world. Whether you are pushing the limits of high-performance AI vector search with Postgres on Silk or scaling Postgres AI workloads even further with Google AlloyDB, the result is the same: Enterprise-grade predictability at extreme scale. Silk eliminates the need to overprovision EC2 compute just to get more storage performance. It eliminates the need to rely on fragile read replicas for your core data path. It gives you the freedom to run your AI workloads on AWS with the exact same enterprise data services and performance guarantees.  Stop praying and start engineering  The AI inference tsunami is already here. The systems that survive it will be the ones built on modern, software-defined cloud storage architectures designed for violent concurrency, massive throughput, and uncompromising tail latency. Don't wait for your own "success disaster" to realize your AWS storage is the bottleneck. It's time to look under the hood and see what an AI-ready data platform looks like. Ready to see the proof? Hear from Eduardo Kassner, chief data & AI officer at Microsoft, and Tom O'Neill, VP of product at Silk, on why AI inference is reshaping system behaviors and why the solution isn't simply adding replicas, adopting new storage systems, or rewriting applications. Watch the webinar now: AI Inference Didn’t Break Your Architecture - It Reveals What Comes Next. Contributed by Silk. ]]></content:encoded>
                <enclosure url="https://image.theregister.com/?imageId=232439&amp;width=800" type="image/jpeg" />
                <media:thumbnail url="https://image.theregister.com/?imageId=232439&amp;width=800" />
            </item>
    <item>
        <guid isPermaLink="true">https://www.theregister.com/a/5225326</guid>
        <link>https://www.theregister.com/software/2026/05/04/how-teamviewer-one-transforms-it-operations-firefighting-aut/5225326</link>
        <pubDate>Mon, 04 May 2026 10:00:30 +0200</pubDate>
        <title>How TeamViewer ONE transforms IT operations from firefighting to autopilot</title>
        <description><![CDATA[ Forget "have you tried turning it off and on again?" Agentic AI support systems now seek and destroy tech issues before they're a problem. ]]></description>
        <category>software</category>
                <lab:kicker><![CDATA[ AI + ML ]]></lab:kicker>
                <dc:modified>Mon, 11 May 2026 10:04:54 +0000</dc:modified>
                <content:encoded><![CDATA[ SPONSORED FEATURE Most IT teams know how quickly they can fix things when they break. This means they can usually tell you their mean time to resolution (MTTR) for helpdesk tickets down to the minute. But MTTR is only part of the picture in complex modern workplace environments. Organizations are more distributed than ever due to hybrid working and ongoing cloud adoption. Compliance frameworks can change almost quarterly, and system complexity continues to mount. IT support teams must shift focus to deal with these situations. "Tickets only provide part of the story and are lagging indicators of performance, addressing issues only after disruption occurs," explains Mark Banfield, chief revenue officer at digital workplace platform provider TeamViewer. "This means there's only limited end-to-end [network] visibility, making it difficult to detect early signals or connect problems across more complex environments." IT support operations have traditionally been reactive, but agentic AI enables them to flip the switch, reducing one of the biggest barriers to productive work: digital friction. This occurs when technology doesn’t work as it should. Every failed authorization or connectivity hiccup reduces productivity and degrades the user experience. A support operation that spots the causes of digital friction and fixes them before them become a problem helps to keep productivity high. No human IT support team can physically keep their eye on thousands of endpoints, including laptops or servers, on a continuous basis. This is where agentic AI can help, monitoring these thousands of endpoints in order to identify patterns that point to emerging issues that might affect users, and intervene early. This means the tech support reporting model can move from reactive to proactive, Banfield says.  The impact of digital friction  TeamViewer's research report, The Impact of Digital Friction, surveyed 4,200 managers and employees worldwide to reveal the implications. It showed that four out of five respondents had lost valuable time to dysfunctional IT, representing an average of 1.3 workdays per month. Just under half (48 percent) indicated the scenario had led to delays in critical operations or projects over the last year. Forty-two percent said such circumstances directly hit revenues, while a further 37 percent said their organization had lost customers over it. To make matters worse, employee experience suffers too. Some 47 percent of those questioned stated that digital friction left them frustrated and less satisfied with their job. A further 42 percent linked it to burnout, while 28 percent said they had considered leaving the organization as a direct outcome. "The result is a growing gap between perceived and actual performance, where unresolved digital friction silently erodes productivity, employee satisfaction, and even revenue," Banfield says. "As environments become more complex and hybrid work scales, reactive models create mounting operational strain and inconsistent experiences." This situation is no longer sustainable, he believes. It's time for organizations to embrace proactive, experience-led IT. Those that don't risk falling behind in both efficiency and competitiveness, he adds.  The shift toward autonomous remediation systems  Real-time insights into what is happening on the network is becoming essential to serve this need, and that makes always-on monitoring systems an imperative. These systems can spot early signs of underlying problems and remediate them autonomously, resolving issues before employees feel the impact. But moving to a truly autonomous system requires a phased approach. It entails moving from reactive support to proactive monitoring, and from there to autonomous remediation. Doing it in phases enables you to regain control of your infrastructure before you move on to more advanced steps, as TeamViewer explains in its playbook on the process. The first stage involves stabilizing your operations by introducing proactive monitoring based on pre-defined policies. This puts you in a better position to understand what has been causing the business problems in the first place so you can fix them. It also makes it possible to see any common patterns of activity so you can automate the response. "Organizations often get stuck between visibility and action, where insights exist but aren't operationalized due to siloed teams or fragmented tools," says Banfield, explaining the importance of this step. "The breakthrough comes when monitoring, automation, and accountability are unified into a closed-loop model that continuously detects, resolves, and improves performance at scale." Proactive monitoring is a great first step. It automates responses to known problems. But it can't prevent issues that haven't been tracked or that support teams are unaware of. This is where agentic AI-based systems like TeamViewer ONE come in. They are not restricted to monitoring the network based on pre-defined parameters or responding to alerts. Instead, these platforms learn from what is happening day-to-day with your most-used devices and apps, such as Microsoft Connect or SAP. They do this by scrutinizing how the environment behaves in real-time, how problems evolve over time, and how support teams intervene. This enables them to recognize which conditions lead to system failure. They also get to understand evolving risks and use this information to automate troubleshooting activities.  The benefits of a continuous feedback loop  "TeamViewer ONE continuously detects issues, prioritizes the ones that matter, and remediates them before users can be impacted," says Banfield. It identifies early signals of digital friction and resolves problems autonomously in real-time before they develop into major issues. In fact, it is the only platform on the market that combines AI with endpoint management, remote access, and digital employee experience (DEX) functionality. The DEX technology, which came from its acquisition of 1E, uses an agent to scan endpoints proactively and also gathers user sentiment about their experience. "Having intelligence on the endpoint is critical because it's the closest point to where issues originate and where employee experience is directly felt," Banfield says. "By embedding visibility, analytics, and autonomous remediation directly at the endpoint, organizations can detect and resolve issues in real time." The result is a continuous feedback loop where known problems are automated out, ensuring they do not happen again. This reduces both ticket volumes and manual effort. The system also documents problems automatically, saving support teams a further step. IT support teams move from reacting to incidents after they occur to resolving problems before users are even aware there were any. This frees up their time to focus on more strategic work, including optimization, and innovation. It also cuts administration costs while supporting security and compliance. A secure, stable environment for employees helps to create a better customer experience, Banfield says. "You can't create a Michelin-star customer experience unless you internally have a Michelin-star employee experience," he explains. "When your employees are benefiting from a proactive IT approach, it will be felt externally by customers as it helps free up time for employees to build a better customer experience strategy."  Enhancing the digital employee experience  TeamViewer points to a US-based home, auto, and life insurance provider as an example of this transition from reactive to automated proactive support. Its IT team had been unaware of just how problematic recurrent issues such as software crashes were becoming for the wider workforce. While many employees just put up with the situation, it was eating into productivity and user satisfaction levels. The team wanted to stop the rot, increasing operational efficiency and enhancing incident resolution times. It rolled out TeamViewer's DEX Intelligence module, which includes an Intelligent Insights AI analytics solution to map trends in IT operations and recommend actionable steps for improvement. This identifies how many devices and employees were being affected by problems, and crystallized a clear picture of end-user frustration for the IT team. With the help of TeamViewer DEX, the tech team found that power management device drivers had been given an "unsigned" status in the company's Windows applications. This made them unstable, causing crashes. It also opened the company up to potential security risks further down the line. Intelligent Insights diagnosed and remediated the issue quickly. Its recommendations and remediation guidance have also helped the IT team make better data-driven decisions to address other red flags before they impact either users or business operations.  Improving customer experience  Another organization that has transformed its endpoint management using TeamViewer's platform is a UK-based global quality food, homewares and clothing retailer. It implemented TeamViewer DEX in 2020 to monitor and manage its complex point-of-sale (POS) systems, plus other endpoint devices, of which there are more than 26,000. Ensuring uptime was a particularly pressing priority for the retailer's POS systems. This is because if they are slow or fail to work, customers are likely to abandon their purchase, which directly impacts sales. But by introducing TeamViewer's DEX platform, the retailer can now pinpoint where problems occur. This has boosted uptime from 92 percent to 98 percent. It has also enabled engineering teams to better monitor, respond to, and predict issues across the estate. This has improved both the employee and customer experience. "Without the disruption of daily friction, employees have a clear pathway to work at their most efficient and productive," concludes Banfield. "There's no lag. It's about fixing issues before they're felt across the organization, ensuring a more consistent, seamless digital experience at scale."  Sponsored by TeamViewer. ]]></content:encoded>
                <enclosure url="https://image.theregister.com/?imageId=254481&amp;width=800" type="image/jpeg" />
                <media:thumbnail url="https://image.theregister.com/?imageId=254481&amp;width=800" />
            </item>
    <item>
        <guid isPermaLink="true">https://www.theregister.com/a/5226977</guid>
        <link>https://www.theregister.com/software/2026/05/04/microsoft-promises-to-do-better-but-it-has-a-long-way-to-go/5226977</link>
        <pubDate>Mon, 04 May 2026 01:00:07 +0200</pubDate>
        <title>Microsoft's turned Windows into a cesspool, but it wants to do better</title>
        <description><![CDATA[ Windows is a mess, GitHub keeps wobbling, Copilot draws flak - what’s wrong at Redmond? ]]></description>
        <category>software</category>
                <lab:kicker><![CDATA[ OSes ]]></lab:kicker>
                <dc:modified>Mon, 04 May 2026 20:31:58 +0000</dc:modified>
                <content:encoded><![CDATA[ KETTLE When it comes to making decisions that piss off your user base, no one knows how to do it like Microsoft.  Relentless Copilot pushing, the second-chance out-of-box experience, a bunch of sloppy, buggy patches, and other bad decisions have plagued the Windows maker in recent months. CEO Satya Nadella and Windows boss Pavan Davuluri have promised to make good, but will they?  This week's Kettle sees host Brandon Vigliarolo joined by US editor Avram Piltch and Microsoft reporter Richard Speed to talk about what's going wrong at Microsoft, and whether the company has lost the thread.  You can listen to The Kettle here, as well as on Spotify and Apple Music. ® Full transcript below: Brandon Vigliarolo (00:00) Welcome to another episode of the Registers Kettle Podcast. I'm Reg reporter Brandon Villarrolo. With me this week is US editor Avram Pilch and our expert on all things Microsoft, Richard Speed. And given we have the Microsoft man with us this week, you can probably guess what we're going to be talking about. And that's just what the heck is going on with Windows lately. Guys, thanks for joining me this week. And I'm so glad we could finally get you on from the UK, Richard. Richard Speed (00:23) Thank you, it's nice to be here. Brandon (00:25) Yeah, yeah, thrilled to have you both here. So jumping right in, our topic this week was actually triggered by a story you wrote, Avram, recently on your experience with quite possibly one of the worst Windows 11 features that Redmond has ever dreamed up, and that's the Second Chance Out-of-Box Experience, or the SCOOBE. For those who aren't familiar with that story, or like me, who have abandoned Windows, fled for the hills years ago, give me a rundown what Scooby is and what the problems with it are. Avram D Piltch (00:52) Well, you could say it's more of a Scooby-Doo than a Scooby-Doo. Yes, I'll be here all week. What happens is, and many of our listeners have probably experienced this, they just didn't know it was called SCOOBE, ⁓ is you turn on your computer, you reboot, or you wake up the computer, and you see it say something like, you're almost done setting up your PC. Brandon (00:55) Duh-duh-ch. And clearly you've done this already, right? And you're confused as to why. Avram D Piltch (01:22) Yes. Yes. So your PC has been set up. ⁓ It may have been set up a year ago or more, but you're on Windows 11 and you're getting there almost done setting up your PC. And it guides you through a process where you are invited to subscribe to different Microsoft services. At the very least, will try to get you to subscribe to Xbox Game Pass, which currently here in the US is $14.99. They will invite you to install Office if you don't have it. Some people report, although I haven't seen this recently, trying to get you to use OneDrive, but... Whatever the case, it's trying to get you to use Microsoft services. Yes, it also has to allow it to change your browser settings, although that's for Edge browser, assuming that you use Edge browser. Brandon (02:15) And change default settings too, right? Like I think browser settings and things, yeah. Okay, so it's not trying to reimpose itself as the default browser necessarily. Avram D Piltch (02:30) It did not when I did it. Brandon (02:32) Okay, Microsoft's learned that lesson, I guess. Avram D Piltch (02:34) It did not change the default browser on me. I had it as Chrome and it didn't change it back to Edge when I accepted that. But they use dark patterns to try and get you to do what they want. there are things like on the browser settings one where it user recommended browser settings. There's a big button at the bottom that says user recommended browser settings. There's no button that says skip. There's no button that says don't. There's a switch that says on and off and you could sort of switch it off and then hit the button or something. But like, they don't make it obvious how you opt out of some of these things. And you can't really opt out of SCOOBE. Brandon (03:12) Right, right. Avram D Piltch (03:21) You can't really opt out of SCOOBE either because it will say, remind me again in three days or something, and then it'll come up in three days. Now, there are ways in the operating system to turn off Scooby, but what I wrote about and what's really important is this is more than just an inconvenience. Brandon (03:41) Yeah, yeah, that was the thing that I honed in on really quick was like, this has got to be, you know, for enterprise, in enterprise environments, this is, this is not, this is a disaster potentially. Avram D Piltch (03:50) Yes, it is because they get support calls for it. And at first, when I heard this from IT people, I said, who are the losers that are calling support because they saw this? Don't they know any better? But then I gave it some thought and I said, you know, it actually makes a lot of sense that you would think that this is a support problem because you turn your computer. Brandon (03:52) Yeah. Avram D Piltch (04:15) And it says you're almost done setting up. So you're like, wait a second, I set up, does this mean some other person has gone into my computer, broken into my computer and run setup while I wasn't looking? Brandon (04:26) Yeah, or did IT push an image to my machine accidentally? Like what exactly just happened? You know, yeah. Avram D Piltch (04:32) So if you know that this is to happen, you can more easily deal with it. But a lot of people don't know that that's BS. They take Microsoft at their word that it's setting up your computer. So this leads to an increase in support tickets. It leads to lost productivity. It seems to happen at the worst possible time. I talked to people who say that they were doing presentations for clients. Brandon (04:38) Mm-hmm. Avram D Piltch (04:59) Or trying to get health information, things like that, where it's mission critical that you get your stuff, and somehow they ended up in SCOOBE when they were trying to do mission critical work. So it's really unethical. Brandon (05:16) Yeah, that's... Yeah, the whole thing is just kinda, yeah, yeah, super unethical, super inconvenient. ⁓ The one question I had, and I know you mentioned in the story, here's ways you can go in to turn it off yourself, and here's ways you can change it in group policy at the enterprise level to stop it. But I mean, that just left me with the question, this is actually enabled by default in enterprise environments and images? I mean, I'm kinda mind blown that this isn't just disabled. Avram D Piltch (05:39) Yes, yes, yes, it is enabled by default. I heard from a number of IT people who they've had to turn it off at the group policy level. So I can't guarantee that this works because Microsoft changes things around a lot. And I wasn't able, I don't have a fleet of PCs to test it on. But supposedly if you go into Group Policy Editor and you go to ... Brandon (05:55) Mm-hmm. Avram D Piltch (06:07) Computer configuration, administrative templates, Windows components, cloud content, and enable turn off Microsoft consumer experiences. That may do it, but that's definitely not off by default. Brandon (06:19) That'll do it. Yeah, well, I guess if you have a fleet of Windows machines, it might not be a good idea to go and update your images just to get this turn off by default when you're deploying your machines. Yeah, exactly. Avram D Piltch (06:32) Because you don't know when it's going to pop up and it definitely results in support calls as if you think about it, it makes sense. Brandon (06:42) Yeah, absolutely. know, people are getting scared. It's popping up. I mean, they don't know what to do. Avram D Piltch (06:46) The other thing that it results in is people doing what Microsoft asked, which the IT department might not want, right? Maybe the IT department doesn't want you running Xbox Game Pass on your company PC. Brandon (06:53) Right, right, yeah. Right? mean, hopefully there would be a billing issue before that would pop up, but you never know. Now, I know that SCOOBE is often the result of Windows machines restarting after an update, right? And they're retrying to get you to, hey, by the way, let's make sure we get all these things back in front of you. But that isn't the only kind of update-related source bot for Windows users recently. And I guess, Richard, you've written several times in the past few months about bad Windows updates. Out of band patches that Microsoft promised are only going to be occasional, becoming frequent. I mean, it looks like there's just kind of a quality control issue with Windows updates recently too, right? Richard Speed (07:37) Yeah, I mean, I would say that definitely is an issue. And the SCOOBE is an example of an update that has probably actually worked. And it seems that seems to be almost an outlier these days. There seems to be almost every month, Microsoft will release a major update. they do major updates on a monthly cadence. You have the Patch Tuesday update, you have the preview update. And normally that should be it. And they contain a bunch of fixes and some new enhancements and what have you. And then if there's any major problem, they'll do what they call an out of band. Brandon (07:55) Mm-hmm. Richard Speed (08:03) Update in between these updates, which contains fixes and what we've been seeing certainly in the last few months is these out of band updates which should have been uncommon - when they're called out-of-band for reason, they are atypical - are happening an awful lot and that speaks. I think of a lot of a lot to. I mean, I can only say it's poor quality control for Microsoft. Not quite sure why it's got as bad as it has, but it's got so bad that I'm obviously. Some leaders within Microsoft have had to come out and say yeah, we're aware of this. I'm going to fix it and but the problem is when these things happen. I mean, I mean a recent one was for example when you try to log in with your Microsoft account, it could respond by saying I'm sorry you've got no internet, which is of course untrue. That was March, yeah, and that's just one example. Brandon (08:36) Yeah. And that was March, that one happened, right? Was that the March? Yeah, and then this month, yeah, and then this month you just wrote about something in the April preview, right? ⁓ Richard Speed (08:56) With April preview they've done their release of patches and so we're keeping our fingers crossed nothing's going to go wrong. Brandon (09:03) Okay, so you've seen, so the April preview problems that you saw have hopefully been fixed. Good. Richard Speed (09:07) Yeah, so the preview we saw from the March preview haven't happened. They released the April preview on April 30th. And it's actually quite an impressive list of fixes because they are addressing stability. They are addressing reliability. There are some useful features in there, which is good, but that's only part of the story because of course what we need to see them now is demonstrate that these features and these fixes that they are rolling out are going to be reliable. not going to see within a week or so another out of band patch to fix a thing that they fixed before, which is what we don't want to see. Brandon (09:39) I guess they have what, 29 days, 30 days to prove that this patch is good before they push the next one. So yeah, we've got a window. Yeah, we've got a window here to see if this actually works. But so at least a couple of bad patches in recent months, ⁓ but it's more than two. Yeah, so a handful of bad patches in the past few months, but that's not it. Bad patches aren't the only thing that's been ⁓ happening with, not the only patch problem, I guess Avram. Richard Speed (09:46) I do hope so. More than two, yeah. Brandon (10:05) You have written more than one Microsoft excoriating opinion piece in the past month. And the second one, I think it was your first one actually, was...how abysmal the update process is for people, especially on machines that may not have a user logging in regularly, right? So in your experience, I think it was just one you have as a spare, but I mean, anyone who's ever run an IT environment with a bunch of Windows machines sitting in a back room somewhere that runs a piece of software that rarely needs human interaction can probably speak to this problem. And that's just...Update deluge. Yeah. Avram D Piltch (10:33) So it's it's punitive. I feel that it's punitive to you as a user if you're not, if you want to update a PC and it's not a PC that you've been using on a regular basis. I'll put in this basket also brand new PCs, right? ⁓ It so happens that at the time of this recording, We just bought my mother a new PC and I need to set it up for her. And the part that I am dreading is all of the Windows updates that we're going to have to go through because it'll probably do several rounds of updates, which is what's frustrating, right? Now, some people might not care and they might just say like, well, I don't care if my machine's updated. And it'll update in the background and eventually I'll reboot it. But I feel like this is kind of part of the setup process. Just get me to the latest version of everything that I need ⁓ so that we're ready to go. And what I find, and I had that issue with a laptop that I kept in a drawer, I needed to use it for some benchmarking and it just seems wrong to benchmark something with not the latest version of Windows on it. So I took it out of the drawer and it took like three hours, like three or four hours and three or four reboots to do the updates on it. And what's frustrating is you think, okay, it's downloaded these updates now, it's changed the build number even to a new build number, it's rebooted, I should be done. Brandon (11:55) Right. Avram D Piltch (12:21) But then you go into Windows settings updates and you click the button to check for updates and it finds a whole new slew of updates that it didn't find the first time you checked. So, I find that super frustrating because I feel like what Microsoft should do is it should get all of the updates that you need to be current in one batch. should get, it should need to- Brandon (12:27) Surprise. Richard Speed (12:32) It's a gift that keeps on giving. Avram D Piltch (12:49) -install a bunch to then go figure out that it needs to install another bunch. Take you through an entire evolution of, oh, you're on this build, we'll take you to that build, then we'll take you to this build. And the advice that I've gotten from some people, and this is good advice, is that if you have an old computer that's been sitting around and you want to update it and it's not a computer, it's a computer where you physically can do this - reinstall Windows from scratch. You would actually have a better time reinstalling Windows from scratch than you would going through the update process that you have to go through. Brandon (13:26) That's not ideal, right? I mean, no, I think, like I said, I kind of abandoned Windows. I jumped ship a while ago. I game on Linux. I have a Mac for my day-to-day experience. You guys are both still Windows users. Do you use other platforms to compare? Because I'm like, how bad? mean, updates are a pain in the butt on my Mac sometimes. But for the most part, my machine will go, hey, you got an update. Let me know when you want to do it. And that's it, right? Richard Speed (13:52) Yeah, yeah, I mean, I yes, I use Linux, I've got some Raspberry Pi machines here and some Linux workstations and also an M1 Mac mini, which I use as well. And I have to say, I mean, I don't know what Avram's experience is, but the Windows Update experience for me is frankly painful. Just, you know, if you don't use a machine as everyone says, it just goes away. I mean, I'm envious of Avram's three hours because I've seen machines go away for even longer as they just update and update. Avram D Piltch (14:19) ⁓ yeah, I mean, Brandon (14:21) Yeah, back in the windows, yeah. Avram D Piltch (14:21) I had that in a drawer for only six, for only a few months. One time, I used to work at another company and they gave me a company laptop and the company laptop was absolute garbage. ⁓ So I used to just use my own laptop every day. And one day they kind of threatened, that like you better have your company laptop on or we're going to check. This was during the pandemic. So I took out the company laptop that hadn't been turned on in three or four years. Okay. And I, and I want, and I tried to let it run Windows updates and it was going good like eight or nine hours to get current. So Brandon (15:06) Absolutely. So, yeah, that's just another example of maybe not the best patch management. Maybe we should work on condensing these a bit, like you said, Avram. But either way, Microsoft has actually heard the cries of the people who are suffering in Patch Valley. And they've come to the rescue, Richard, with a solution that is probably not the best one, right? Richard Speed (15:30) Yeah, it's a tricky one. mean, the solution that's come out has been because you've always been able to pause updates and you've always been able to kind of specify when you want them to install. I don't want to say always, they've made changes over the years. The latest changes, you can now pause them pretty much forever. I think it's, mean, correct me if I'm wrong, it's a 35 day pause. can say, okay, remind me again in 35 days and you can keep doing that over and over again. Now that initially sounds like quite a good thing. Brandon (15:43) Right. Richard Speed (15:56) However, I'd argue it's actually quite a bad thing. And that's because, I mean, we're talking about updates here and the pain they cause, but they are a good thing. mean, these updates, this is one of reasons why I find the out-of-band patches and people worrying about postponing them concerning because often they contain fixes for vulnerabilities and you don't want to postpone those. You want those fixes in place. And so the ability that Microsoft is rolling out now to effectively indefinitely pause a patch, that actually concerns me because it's like, well, you do want these figures in place. So I think it will be interesting to see how that plays out and how that's actually deployed. if that can be overridden so highly critical patches for vulnerabilities get rolled out without a pause. Brandon (16:42) Right. Yeah, that would be great, right? And I mean, ideally that would happen in the background. I mean, it's funny, we're in an era where like, you know, I play a lot of online gaming, lot of online games, and there's games nowadays that can perform updates to persistent worlds, right, without having to shut down or kick anybody off. And it's like... Richard Speed (16:58) Well, yes, mean, there are games that don't. I'm thinking Microsoft Flight Simulator, for example, which last time I tried to play that, that went away for about three hours beforehand to an update. And you just want to pick up the game and play. Brandon (17:04) Uh-huh. Well, you did utter the magic word before you said the name of the game, though, and that was Microsoft. I can't imagine. But, know, so it looks like maybe Microsoft's patching experience isn't great. Hopefully they'll make some fixes. But ⁓ Richard, you've covered it. of fixes and improving the experience, know, ⁓ Windows Boss Pavan Dabaluri has made a lot of promises recently, and even ⁓ Satya Nadella has said this latest patch that just came out was kind of their attempt to win back frustrated users. mean, you know, is this showing signs of actual improvement here, this patch, and the promises? Richard Speed (17:36) I would say yes and no. Yes, that what [Windows and Devices Executive VP] Pavan [Davuluri] has said, I he made some statements in last November, he made some more statements in March. And there's an acknowledgement that okay, Windows has become a bit unreliable, it's a bit unstable, they're gonna work on that. They're aware that they perhaps sprayed copilot in too many places without really thinking too hard about it. So that's good news. I mean, I have to say, I did notice certainly in Pavan's statements, I didn't see the word sorry or apology in there at all which well exactly because - Brandon (18:13) Yeah, yeah, I made a note of that too that I wanted to hit on that particularly, right? Like it was very, one of those carefully worded, you know, things, yeah. Richard Speed (18:20) Exactly. This is case you think, well, you know, I think you owe the windows enthusiast community an apology because it has definitely got worse. And I think just this week when Microsoft announced its financials, we had its CEO, Sacha Nadella, saying, you know, they were putting foundations in place to win back the fans that they've lost. And the question there is, well, how did you lose these fans? What's, you what's happened? I think we all know what's happened. The quality has gone down. They have stopped listening so much. But it isn't a new thing. I the quality issues within Windows have been there for many years. It just appears to be getting particularly bad now. Brandon (18:55) Yeah, yeah, because I was going to say one of the things I wanted to mention was that you said was, you know, like the note of we're going to be more purposeful with where we put Copilot, right? And in the thought, the response that you had to that was, well, does that mean you just haven't been being purposeful up to this point? And it's one of those things that, like, you know, it's kind of shocking to hear some of the ways they won't apologize for things. But there is still the tacit admission that they did something wrong, you know. Richard Speed (19:10) What exactly? Yeah, which. Yeah, but I think you can compare and contrast that to say one of the companies that Microsoft owns, GitHub, which has apologized and said it's sorry for what's happened. Brandon (19:30) Yeah, I was actually gonna say that. Yep, because I was gonna move into that and talk more broadly actually. We were talking about Windows specifically, but GitHub specifically made me think of more broadly outside of Windows. Microsoft isn't just having problems in the Windows front. I think the GitHub issue recently is largely due to stability. It's the platform's not been reliable. Things have been going down to the point where people are leaving and saying that it's time to abandon GitHub and flee for the hills. You know, Microsoft, I think, has blamed AI eating up resources for lot of that outage, I believe, in articles you've written. Is that correct, Richard? Richard Speed (20:05) Yes, I mean they've certainly pointed the finger blame at add AI. mean, GitHub. Obviously you've got the GitHub copilot which would use AI to generate code for you. Brandon (20:09) Yeah, well, I was gonna say that it seems like, know, again, right, like Microsoft is kind of shooting themselves in the foot, right? They've been trying to... We didn't touch on Copilot too much in this conversation because there are so many other problems with Windows to talk about. But, you know, they were shoving Copilot everywhere they could to the point where people were like, why are you... We don't need it here, you know? And it has been really integrated into GitHub. And Microsoft's pushing that AI hard. Use this GitHub CLI, use these agents, blah, blah, blah, blah. But now they're telling us that, the platform we created that you're supposed to integrate all these agents is failing because the push that we made, what's that? exactly, right? And that's kind of like, it makes me wonder more broadly beyond Windows, like does Microsoft have, they just seem to continue to fail to meet user needs and expectations. Avram D Piltch (20:55) Whose fault is that? Whose fault is that? Brandon (21:10) And now instead of, think both of you have noted this in your stories, instead of fixing their problems, they simply are jumping ahead to the next thing. You know, they've got problems in Windows, they've got problems with AI, they've got problems with GitHub, right? Has Microsoft lost the focus? Have they lost their way? I mean, I'm just kind of curious. You know, you guys are both who know Microsoft. What do you think? Avram D Piltch (21:29) They're in the way of, they're in the zeitgeist, right? Everybody's doing AI, Microsoft has to do AI. And the idea is to show the investors and to show the press, ⁓ we are more AI than anybody else. What I would like to see Microsoft do is focus on productivity for the human, make the human more productive. Over the summer, I did a story about some of the things I'd actually like to see them do in a positive way. And they're never focused on how do we actually make you more productive? They're trying to focus on how do we make you offload your work to a bot? But they could do things in Windows that would actually make it better. Like give me multiple clipboards so I can copy and paste multiple things. You know. Give me a firewall for audio so I can decide. So if something's about to play a sound, I can stop it. Give me, you know, give me an extra modifier key so that I can create more keyboard shortcuts. Like there's all kinds of things that Microsoft could do in Windows to really help people be more productive if they were focused on. Richard Speed (22:46) Yeah, I and I'll add to that, because I'm using the I word, which is investor. I think Microsoft has become way too focused on returning value or creating value for investors....Microsoft has lost touch with consumers. It's a big company. It's not so nimble now, but I do think it stopped listening. It stopped listening to the things that Avram has said just now. It stopped listening to what consumers are wanting. So I'm hopeful that maybe we'll see it start doing that. I mean, I think it's going to have to because at the moment it's a massive company, but it depends on a pipeline of enthusiasts who are going to come through. And I think this is why I mentioned winning back fans, because once you lose your enthusiasts and people just regard it as yet another tool for work, or just another thing that's in the office, then in a few years time, those same people who perhaps anything with Microsoft, they're going to start becoming the decisionmaker in enterprises. And that's where think Microsoft could come unstuck. It's going to take a few years, but I think it's recognized it needs to start paying attention to these enthusiasts again, where I think it may have lost its focus in last few years as it focused purely on raising value for investors through AI buildouts. Brandon (24:07) Yeah, classic enshitification and now it's kind of that moment of, wait, did we, oops, you know, and so maybe they've got to rethink, but that'd be great. I would love it if Microsoft was actually taking a moment to pause and go, "right, it's the consumers who we're building this for. They're the ones actually spending the money, you know?" Richard Speed (24:14) Yeah. Well, exactly. Yeah, and what's what happens? mean, I mean, we shouldn't lose sight of the fact that obviously, you know, it is building out a lot of AI stuff. AI is a great system. Don't get me wrong. AI is a great tool. I think occasionally it's been it's been pitched as perhaps doing too many things. And I think Microsoft's focus on that means it's lost focus elsewhere. And I think is kind of what the problem is. And we're seeing it in the drop in quality. We're seeing it. But also we're seeing -- you use the word enshittification The SCOOBE is a great example of why are they doing that awful thing? I mean, I don't know about you, but I've had I've I mean Avram D Piltch (24:58) It's obvious why they're doing it. mean, somebody has a quota to me of subscriptions and they have a captive audience and they're going to use it. And Windows 11 is designed to push what Microsoft wants, not to give you what you want. And what's really insidious about that is that you paid for it. is people do not pay for, you know - the average end user does not pay for Linux. The average end user Mac OS comes with your Mac. So Microsoft is charging you for an OS and then using it to sell you things. Brandon (25:36) Well yeah, the key to the future is recurring revenue, right Avram? You gotta get that from somewhere. yeah, that's a very good point too. Either make it free or make it cost once, right? Avram D Piltch (25:40) That it should be free. If you're spending money, you shouldn't get asked. Brandon (25:50) Yeah, I agree. Well, you know, we'll see what comes. I Microsoft's got some rebuilding to do. Maybe they're in that rebuilding year, rebuilding phase here where they're trying to get back to a good starting lineup. We'll see what comes of it, right? And you know, no matter what way it shakes down, we are going to probably be right here on the kettle to talk about it. And we would love to see you back when we do. Thanks for coming out this week, guys. Richard Speed (26:12) Thank you. Avram D Piltch (26:13) Thank you. ]]></content:encoded>
                <enclosure url="https://image.theregister.com/?imageId=239474&amp;width=800" type="image/jpeg" />
                <media:thumbnail url="https://image.theregister.com/?imageId=239474&amp;width=800" />
            </item>
    <item>
        <guid isPermaLink="true">https://www.theregister.com/a/5229422</guid>
        <link>https://www.theregister.com/software/2026/05/03/inference-is-giving-ai-chip-startups-a-2nd-chance-to-shine/5229422</link>
        <pubDate>Sun, 03 May 2026 15:05:11 +0200</pubDate>
        <title>Inference is giving AI chip startups a second chance to make their mark</title>
        <description><![CDATA[ In a disaggregated AI world, Nvidia can be both a friend and an enemy ]]></description>
        <category>software</category>
                <lab:kicker><![CDATA[ AI + ML ]]></lab:kicker>
                <dc:modified>Fri, 01 May 2026 12:05:15 +0000</dc:modified>
                <content:encoded><![CDATA[ AI adoption is reaching an inflection point as the focus shifts from training new models to serving them. For the AI startups vying for a slice of Nvidia's pie, it's now or never. Compared to training, inference is a much more diverse workload, which presents an opportunity for chip startups to carve out a niche for themselves. Large batch inference requires a different mix of compute, memory, and bandwidth than an AI assistant or code agent. Because of this, inference has become increasingly heterogeneous, certain aspects of which may be better suited to GPUs and other more specialized hardware.  Nvidia's $20 billion acquihire of Groq back in December is a prime example. The startup's SRAM-heavy chip architecture meant that, with enough of them, Groq's LPUs could churn out tokens faster than any GPU. However, their limited compute capacity and aging chip tech meant they couldn't scale all that efficiently. Nvidia side stepped this problem by moving the compute heavy prefill bit of the inference pipeline to its GPUs while it kept the bandwidth-constrained decode operations on its shiny new LPUs.  This combination isn't unique to Nvidia. The week after GTC, AWS announced a disaggregated compute platform of its own that used its custom Trainium accelerators for prefill and Cerebras Systems' dinner-plate sized wafer-scale accelerators for decode. Even Intel has gotten in on the fun, announcing a reference design that'll use GPUs — presumably the one they teased last northern hemisphere fall — for prefill and AI chip startup SambaNova's new RDUs for decode. So far, most of the AI chip startups' wins have been on the decode side of the equation. SRAM, while not particularly capacious, is stupendously fast. So with enough chips, or at least a big enough chip in the case of Cerebras, they're well suited to accelerating decode operations, but chip startups aren't limited to this regime. This week, Lumai detailed its optical inference accelerator, which uses light, rather than electrons, to perform the matrix multiplication operations at the heart of most machine learning workloads using a fraction of the power of a purely digital architecture. Lumai expects its next-gen Iris Tetra systems will achieve an exaOPS of AI performance in a 10kW power budget by 2029. Technically, the chips use hybrid electro-optical architecture, but the bulk of the compute done during inference is handled by the chip's optical tensor core. Initially, the company is positioning the chip as a standalone alternative to GPUs for compute-bound inference workloads, such as batch processing. Longer-term, the company also plans to use its optical accelerators as prefill processors. The architecture is still in its infancy, capable of running billion parameter models like Llama 3.1 8B or 70B today, but it's far enough along that the UK-based startup has opened its chips up to neoclouds and hyperscalers for evaluation.  Having said that, not every AI chip startup is keen on using different chips for prefill and decode. Earlier this week Tenstorrent unveiled its RISC-V-based Galaxy Blackhole compute platforms, and suffice to say the company's CEO Jim Keller isn't a fan of the disaggregated inference formula.  "Every company in the industry is pairing up to build the accelerator accelerator accelerator. CPUs run code. GPUs accelerate CPUs. TPUs accelerate GPUs. LPUs accelerate TPUs. And so on. This leads to complex solutions which are unlikely to be compatible with changes in AI models and uses. At Tenstorrent, we thought something more general and simpler would work," he said in a statement. ® ]]></content:encoded>
                <enclosure url="https://image.theregister.com/?imageId=225583&amp;width=800" type="image/jpeg" />
                <media:thumbnail url="https://image.theregister.com/?imageId=225583&amp;width=800" />
            </item>
    <item>
        <guid isPermaLink="true">https://www.theregister.com/a/5230018</guid>
        <link>https://www.theregister.com/software/2026/05/02/how-to-roll-your-own-local-ai-coding-agents/5230018</link>
        <pubDate>Sat, 02 May 2026 13:30:07 +0200</pubDate>
        <title>Usage-based pricing killing your vibe - here's how to roll your own local AI coding agents</title>
        <description><![CDATA[ Take those token limits and shove them by vibe coding with a local LLM ]]></description>
        <category>software</category>
                <lab:kicker><![CDATA[ AI + ML ]]></lab:kicker>
                <dc:modified>Fri, 01 May 2026 20:25:26 +0000</dc:modified>
                <content:encoded><![CDATA[ With model devs pushing more aggressive rate limits, raising prices, or even abandoning subscriptions for usage-based pricing, that vibe-coded hobby project is about to get a whole lot more expensive. Fortunately, you're not without cost-saving options. Over the past few weeks, we've seen Anthropic toy with dropping Claude Code from its most affordable plans while Microsoft has skipped testing the waters and moved GitHub Copilot to a purely usage-based model. The whole debacle got us thinking. Do we even need Anthropic or OpenAI's top models, or can we get away with a smaller local model? Sure, it might be slower, less capable, and a little more frustrating to work with, but you can't beat the price of free... Well, assuming you've already got the hardware that is. It just so happens that Alibaba recently dropped Qwen3.6-27B, which the cloud and e-commerce giant boasts packs "flagship coding power" into a package small enough to run on a 32 GB M-series Mac or 24 GB GPU.  What's changed  This isn't the first time we've looked at local code assistants. Previously we explored using Continue's VS Code extension for tasks such as code completion and generation. At the time, the models and software stack were quite immature, making them useful tools, but not necessarily good enough to compete with larger frontier models. Since then, model architectures and agent harnesses have improved dramatically. "Reasoning" capabilities allow small models to make up for their size by "thinking" for longer, mixture-of-experts models mean you don’t need terabytes a second of memory bandwidth for an interactive experience, and vastly improved function and tool calling capabilities mean that these models can actually interact with code bases, shell environments, and the web.  All vibes, no rate limits  In this hands on, we'll be looking at how to deploy and configure local models like Qwen3.6-27B, for coding on your computer, and explore some of the agent frameworks you can use with them. What you'll need:  A machine capable of running medium-sized LLMs. We recommend an Nvidia, AMD or Intel GPU with at least 24 GB of VRAM. If you're a little short on memory, we'll also discuss how to pool your system and GPU memory. For those on newer Mx-Max series Macs, we recommend at least 32 GB of unified memory. For this guide, we'll be using Llama.cpp to run our model, but if you prefer to use LM Studio, Ollama, or MLX, the set up process is similar. If you need help getting Llama.cpp installed on your system, you can find our comprehensive setup guide here.  Note: Older M-series Macs may struggle with the large context lengths required for agentic coding. You may have better luck with an inference engine like oMLX, which can take better advantage of Apple's hardware accelerators, but your mileage may vary.  Spinning up the model  Running LLMs locally is a dead simple process these days. Install your favorite inference engine. Download the model, and connect your app via the API. However, for code assistants in particular, there are a couple of parameters we need to dial in, otherwise the model is apt to churn out garbage and broken code. Some models require specific hyper-parameters to function properly in different applications, and Qwen3.6-27B is no exception. When using Qwen3.6-27B for vibe coding, Alibaba recommends setting the following parameters:  temperature=0.6 top_p=0.95 top_k=20 min_p=0.0 presence_penalty=0.0 repetition_penalty=1.0  We also need to set the model's context window as large as we can fit in memory. If you're not familiar, a model's context window defines how many tokens the model can keep track of for any given request. When working with large code bases containing thousands of lines of code, this adds up quickly. What's more, the system prompts used by many agent frameworks can be quite large, so we want to set our context window as high as possible. Qwen3.6-27B supports a 262,144 token context window, but unless you have a high-end Mac or a workstation GPU, you probably don't have enough memory to take advantage of all of that, at least not at 16-bit precision. The good news is that we don't need to store the key-value caches, which track the model state, at 16-bits. We can get away with lower precisions without too much performance and quality degradation. To maximize our context window, we'll be compressing the key value pairs to 8-bits. Finally, we'll want to make sure prefix caching is turned on. For workloads where large sections of the prompt are going to be reprocessed over and over again, like a system prompt or code base, this will speed up inference by ensuring only new tokens are processed. In newer builds of Llama.cpp this should be enabled by default, but we'll call those flags just in case. With all that out of the way, here's the launch command we're using for a 24GB Nvidia RTX 3090 TI, but the same code command should work just fine if you're using an AMD or Intel GPU or are running Llama.cpp on a Mac. If you're running this on a machine with more memory, try bumping up the context window to 131,072 or 262,144. If you're planning on running Llama.cpp and accessing it on another machine, you'll also want to add --host 0.0.0.0 to the command, which will expose it to your local area network. If Llama.cpp is running in a VPC, you'll want to configure your firewall rules before passing this flag for the sake of security.  Choosing an agent framework  Now that our model is up and running, we need to connect it to an agentic coding harness. On their own, models can generate code, but they have no way to implement, test, or debug it without an active development environment. Part of what has helped vibe coding take off where other AI ventures have struggled, is that code is verifiable. It either runs or compiles, or it doesn't.  To keep things simple we'll be looking at three popular options: Claude Code, Pi Coding Agent, and Cline. We'll kick things off with Claude Code. Despite what you might think, you don't have to use Claude Code with Anthropic's models. The framework works just fine with local models, assuming you've got enough resources to run them. Install Claude Code as you normally would. You can find Anthropic's one-liner here. Next, we'll need to tell Claude Code we want to use the model running locally on our machine rather than a Claude account or Anthropic's API services. This is done by setting a few shell variables before launching Claude Code. These will need to be run each time you launch Claude from a new session. Now when you start Claude, it'll connect directly to your local model. Claude Code itself continues to function as it normally would.  Pi Coding Agent  Let's say you not only want to use your own local models, but would prefer an open source harness as well. If you like Claude Code, you'll probably like the Pi Coding Agent. And just like Claude Code, it's not picky about what model you use with it. One of the main attractions of Pi Coding Agent is how lightweight it is. Long input sequences can be extremely taxing on lower end or older GPUs or accelerators. Claude Code and Cline both have system prompts that can bring less capable hardware to a crawl. By comparison, Pi Coding Agent's default system prompt is short enough to keep things snappy, especially with prompt-caching enabled. However, that speed comes at the expense of many of the guardrails and safety features we see on other coding agents. This is one you'll probably want to spin up in a virtual machine, container, or even a Raspberry Pi. Much like Claude, the Pi Coding Agent can be installed using the appropriate one liner for your system. After that, all that's required is a little bit of JSON telling the agent harness where to find your model. If you've been following along, the setup is fairly simple. Using your preferred text editor, create the following file: Windows: Linux / Mac: Next, paste in the following template. If you've set an API key, replace no_API_key_required with your key. The rest of these will depend on what model and port you're using. You'll also want to adjust the contextWindowSize to match what you set in Llama.cpp. With that out of the way, we can navigate to our working directory, launch Pi Coding Agent, and get to work vibe coding our next hobby project.  Cline  Claude Code integrates directly with popular integrated development environments (IDEs) like VS Code, but if you're going this route, we also recommend checking out another open source app called Cline. Installing Cline is as simple as finding it in VS Code's — or a supported IDE's — extension manager and adding it to your library. Next, we'll point Cline at our Llama.cpp server and adjust a few hyperparameters like temperature and context size:  Base URL: http://localhost:8080/v1 Model ID: unsloth/Qwen3.6-27B-GGUF:Q4_K_M Context Window Size: 65536 (Or whatever you set in Llama.cpp) Temperature: 0.6  Once it is configured, you can interact with Cline through its chat interface. Any files or edits will appear in VS Code as they're generated. One of Cline's more useful features is the ability to switch between a pure planning mode and an action mode. If you've ever gotten frustrated because Claude interpreted a question as a call to action when what you really want to do is workshop a problem, this is a huge help.  Are local models finally good enough?  So can Qwen3.6-27B replace Opus 4.7 or GPT-5.5? Not exactly. As you probably guessed, a 27B LLM isn't a replacement for a multi-trillion parameter frontier model. However, you might be surprised with just how far you can get with local models these days. In our testing, Qwen3.6-27B easily one shot an interactive solar system web app and was able to accurately identify and patch bugs in an existing code base. Admittedly, these are fairly trivial projects. To get a better sense of how well the model performs, I handed it over to fellow vulture Thomas Claburn to see how it compares to his recent experience with Claude Code. He writes:  Are these agents even safe?  With all the hullabaloo over the security nightmare known as OpenClaw, it's a good question. Thankfully, most of the frameworks we've discussed here are fairly limited in their autonomy. By default, Claude Code, and Cline rely on having a human-in-the-loop to approve code changes and execute shell commands. Unless you've whitelisted a set of commands or are spamming the enter key before reading without taking the time to understand what it is that the agent is trying to do, the blast radius should be manageable. We emphasize "should be" because a basic understanding of the programming language and common CLI commands goes a long way here. If the model starts asking to run rm -rf on files or folders outside your working directory, something probably has gone wrong. This isn't the case with Pi Coding Agent, which operates in YOLO mode out of the box, which gives it free rein to read and modify anything it has access to. In a dedicated development environment like a virtual machine or Raspberry Pi, this might be an acceptable risk, but if it's not, you may want to consider running the agent in a proper sandbox. Containerization offers an easy avenue for this. It's fairly simple to spin up a Docker container and pass your working directory through to it. Docker is a whole can of worms on its own, but the following run command should give you a reasonable starting point for a sandboxed environment. You can find instructions on installing Docker on your preferred OS here. This will spin up a new Ubuntu docker container and pass through our working directory to the container. Any changes will be limited to that folder or the container. If you'd like to see a comprehensive guide on building agent sandboxes, let us know in the comments section. ® ]]></content:encoded>
                <enclosure url="https://image.theregister.com/?imageId=5230298&amp;width=800" type="image/jpeg" />
                <media:thumbnail url="https://image.theregister.com/?imageId=5230298&amp;width=800" />
            </item>
    <item>
        <guid isPermaLink="true">https://www.theregister.com/a/5225216</guid>
        <link>https://www.theregister.com/software/2026/05/01/cios-will-be-the-governors-for-ai-agents/5225216</link>
        <pubDate>Fri, 01 May 2026 16:54:13 +0200</pubDate>
        <title>CIOs ready for another role-change as AI becomes agent of chaos</title>
        <description><![CDATA[ If software writes software the risk is “systematic failure at scale”. Someone needs to take charge, argues Forrester ]]></description>
        <category>software</category>
                <lab:kicker><![CDATA[ AI + ML ]]></lab:kicker>
                <dc:modified>Fri, 01 May 2026 14:53:38 +0000</dc:modified>
                <content:encoded><![CDATA[ Forrester predicts that by decade's end, the rush toward agentic AI will grow so chaotic that CIOs will be forced into a new role as enforcer of order. In a recent research note, the analyst warned that the promise of line-of-business departments building and deploying their own AI agents will fade as agent systems sprawl across the organization, increasingly misaligned with business needs. Forrester said CIOS would end up "governing the enterprise AI-powered operating system" rather than running the tech. This is because the proliferation of AI agent systems, built into application software and cloud infrastructure, could lead to “fragmented adoption, weak data foundations, unclear decision-rights, or incomplete process design." "In 2030, these errors will create systematic failure at scale. The challenge is no longer making transformation stick but ensuring the enterprise doesn’t outrun the capacity for control,” the research paper said. This is the same Forrester which, in October last year, said fewer than one-third of decision-makers were able to make the connection between the value of AI and their corporation's financial growth. It meant large organizations were set to defer a quarter of planned AI spending from 2026 until 2027. Enterprise application vendors are using their entrenched positions among customers to end discounting and push high-margin AI products, Forrester also said. Nonetheless, AI agents will, in the end, come to dominate enterprise IT, it now claims. "As software generates software and autonomous agents execute work, the CIO’s center of gravity shifts from building systems to governing outcomes," the latest paper said. As such, CIOs must don a new cape, change their hats and find their way in the world wearing an entirely new costume fit for roles. Characters they may want to inhabit include: the architect of enterprise decision-making. "Platforms must support real-time decision-making while enforcing constraints that protect margin and reduce downside risk," the paper said. Another part will be the governor of autonomous systems. "CIOs must define bounded autonomy patterns that control how far an agent’s decisions propagate and how quickly the enterprise can intervene." The more ambitious among CIOs might be tempted to play the role of "storytellers who translate probabilistic risk into confidence." That means, showing how uncertainty is "managed, monitored and owned," rather than brushed aside, the paper said. ® ]]></content:encoded>
                <enclosure url="https://image.theregister.com/?imageId=5225709&amp;width=800" type="image/jpeg" />
                <media:thumbnail url="https://image.theregister.com/?imageId=5225709&amp;width=800" />
            </item>
</channel>
</rss>