For only the third time in 25 years, the FAA has called a ground stop for all air traffic in the US. US officials said a telecommunications issue has hampered Air Traffic Control Centers across the US and their ability to communicate with planes in the air.

Around 4:30PM ET, the FAA issued, according to ABC News, an extraordinary alert to pilots that reads in part:

"Due to extraordinary circumstances and for reasons of safety. Attention all aircraft operators. By order of the Federal Aviation Command Center, all airports are ground stopped."

While US Transportation Secretary Sean Duffy has yet to issue a public statement (beyond the initial announcement of the ground stop), Department of Homeland Security Markwayne Mullin told reporters that this grounding, or the cause of it, "is not a threat to the US," and added it was a telecommunications outage.

He did not detail the exact affected technology, though, when asked about the coincidence of the massive Microsoft Outlook outage yesterday and this today, Mullin said, "I don't see any reason to connect those two at all."

A Microsoft X account also made the same point, writing, "We have no indication that the FAA’s nationwide ground stop is related to any Microsoft services."

On the possibility that this is a cyberattack, Mullin said they were not ruling anything out and "As we investigate what caused this, we always have to assume — as many cyberattacks take place —we make that assumption, but that's not known."

The ongoing grounding comes a day after the FAA made an announcement of its AI-backed SMART system (Strategic Management of Airspace Routing Trajectories) to more airports. However, that rollout started back in September. So any connection there would be conjecture.

There was a moment, though, when someone asked Mullin about that AI system, and Mullin quipped, "We call it 'SI' around here."

As of now, air traffic control systems around the country are managing landings on what Mullin called "a case-by-case basis."

We started tracking the skies over the US via Flightradar24 more than an hour ago, and its depiction of in-flight planes and the direction they're heading over the last hour or so is remarkable. It went from being a fairly normal flight day to a visually chaotic race back home. Flights that may have taken off appear to have been rerouted back to their nearest airports.

While JetBlue and American Airlines websites make no mention of the FAA grounding, Delta has a banner at the top of its site and recommends that everyone check their Delta App for their flight status.

For now, Mullin said pilots and air traffic controllers can communicate "at certain levels", but they have some "overwhelmed" communication systems.

Online pundits, including pilots and aviation experts, think there may be a fairly pedestrian issue. Scott Manley, a YouTube science expert and pilot, wrote on X:

"It sounds to me like the ATC outage is limiting the ability for enroute centers to communicate with each other. As aircraft move across the country, they get handed off from one controller to the next, and your flight information is handed between controllers in a separate channel."

Whatever the cause, it may take a day or more to untangle this mess.

Herman Miller is one of the best-known and most respected chair brands. Their office chairs, in particular, are highly coveted and have gained a significant following thanks to their true ergonomic design. Arguably, Herman Miller is one of the most credible names in the world when it comes to chair design.

That's why we're so happy to let you know that the Herman Miller Black Friday sale has launched, giving customers 25% off everything!

Moving quickly from task to task? Mirra 2 might be your chair. Want to make a statement? Give Sayl a chance. Feel like you need your own support system? Go for Aeron. For any chair you choose, you can be confident in decades of research-based ergonomic innovation, 12-year warranty, and continued sustainable stewardship. So however you sit, and whoever you are, you'll be happier in a Herman Miller.

Get 25% off at Herman Miller

Top picks from Herman Miller

Why buy from Herman Miller

Herman Miller is a premium furniture brand that sells office chairs, desks, tables, sofas, and stools. All of their products are known for their high quality and longevity.

They are also built to last, as evidenced by the fact that most of them come with a 12-year warranty. When the cost of the item is spread over its lifetime, the premium price becomes worth every penny. Their products are made with high-end components, which increase the level of durability.

In terms of ergonomics, the Aeron chair is a prime example. It features adjustable PostureFit SL for spinal support, balanced recline, and 8Z Pellicle for weight distribution. Other Herman Miller chairs, like Embody, have unique features like a pixelated support system for flexible seating and a wide range of backrest movement.

If you're in the US and are looking to upgrade what you spend most of your day sitting on, then now is a great time. Herman Miller regularly has live offers, and you can check out our Herman Miller coupon codes hub for more offers throughout the year.

• The us DoJ said it obtained a court order to "freeze or seize" $1.2 billion out of the $6.7bn stolen from Binance
• It also disrupted the infrastructure used in FaceStealer attacks
• Most of the Binance funds remain unaccounted for

The US Department of Justice (DoJ) has said it was granted court authorization to “seize or restrain” $1.2 billion recently stolen from Binance.

In late October 2026, Binance reported losing $6.7 billion in a highly sophisticated heist that played out over roughly 20 minutes. In the days that followed, we learned that the heist was carried out by TraderTraitor, a subset of the infamous Lazarus Group, notorious North Korean state-sponsored threat actors responsible for some of the biggest crypto robberies to ever happen.

The community rallied around Binance, helping trace the funds and label the attackers’ wallets. Now, US law enforcement says at least part of those funds was seized or frozen. The agency did not say what that meant, how it was done, or if Binance will ever see those tokens again.

Almost $3bn seized

“Since the theft, the National Security Division’s National Security Cyber Section, the Criminal Division’s Computer Crime and Intellectual Property Section, the U.S. Attorney’s Office for the District of Columbia, and law enforcement partners have worked closely with virtual asset service providers (VASPs) to interdict the stolen assets,” the DoJ said in the announcement.

“Collective U.S. government and private sector efforts so far have resulted in the freeze of more than $2.7 billion, with the Department having already obtained court-authorization to seize or restrain in excess of $1.2 billion of that amount.”

In the same report, the DOJ also said the US government disrupted cloud-based infrastructure that North Korean hackers used in a social engineering scheme called FaceStealer. This scheme was allegedly not used in the Binance hack, but it did result in the theft of around $400 million over the last 18 months, cumulatively. Furthermore, the methodology is almost identical to what TraderTraitor used in its attack.

FaceStealer

We also learned that the preparation for the attack took months, and that the group broke into Binance through an employee’s device. That employee was allegedly invited to a fake job interview, through which they were served malware that ultimately granted TraderTraitor valuable access. How they escalated that access into the biggest crypto heist ever remains to be seen.

The group pulled the money from 10 different chains, including Bitcoin, Ethereum, XRP, Solana, Arbitrum, and others. Once the money - which belonged to Binance and not its customers - left the company’s wallets, it was sent to different “mixers” and other laundering services.

FaceStealer is a scheme that does exactly what the name suggests - steals people’s faces, so that they can be used in identity theft which is a vital cog in the larger crypto theft engine.

According to the DoJ, criminals would first compromise an account belonging to their target’s contact, and try to get them on a video call. That call would be done on a fake Zoom platform - their victim would see their contact’s video feed which, in reality, was just a previously stolen recording. They would hear no audio, and would then receive a message that they need to update their computer’s driver to be able to hear audio and participate in the call. The driver, in this scenario, is the malware, usually either an infostealer or a steppingstone into a wider network.

North Korean state-sponsored actors are some of the most active and most disruptive hackers around. Multiple sources from the cybersecurity community claim the government is using the stolen money to fund its state apparatus, as well as its nuclear weapons programme.

On the odd occasion the North Korean government responds to these allegations, it usually just dismisses them as baseless and instead accuses the US of doing the same thing.

Via The Register

• Threat actors used 4,837 compromised GitHub accounts to open pull requests across 61,764 repositories
• At least 1,835 repositories executed the payload and had credentials exfiltrated
• Authentication tokens, AWS, Azure and Google Cloud credentials, and more, targeted

Cybercriminals were recently seen abusing multiple GitHub features to deploy infostealers and exfiltrate user’s cloud login credentials and other secrets.

Tens of thousands of repositories have already been targeted, and almost 2,000 fell prey to the attack, cybersecurity researchers Aikido said, in a recent report outlining how the attackers used stolen GitHub credentials to abuse GitHub Actions and workflows, two features designed to speed up and automate certain activities.

GitHub is a platform for developers working on open source projects. Each project has its own “repository” - a folder containing source code, configuration files, and other resources. Developers can propose changes to public repositories through a process called “pull request”, after which other developers can review and approve/decline the changes.

GitHub also has an automation service called Actions, that allows developers to automatically execute various tasks, including pull requests. These automated processes are called workflows and can be described as a set of instructions telling GitHub what to do. Here is another vital nugget of information: workflows can access sensitive credentials when required to perform these tasks.

Using compromised accounts

According to Aikido, the threat actors used 4,837 compromised GitHub accounts to open pull requests across 61,764 repositories. The researchers did not say how the attackers obtained these accounts, or how they were compromised. The pull requests were probes, not actual attacks, and were used so the attackers can see which projects they could modify.

“We're tracking this campaign as wfx-loot based on the threat actor's name for the malware and the term used for stolen secrets,” Aikido said. The pull requests were titled docs: dependency audit notes, and were submitted from branches named wfx-docs.

Of the 60,000+ matching pull requests, more than 10,000 targeted repositories the compromised accounts did not own, but could access. After establishing write access, the attackers used one of three methods to deploy a malicious script: modifying existing GitHub Actions workflows, adding a shell script triggered by an npm installation hook, or creating a new workflow file on a temporary branch.

The payload included roughly 1,500 lines of code that was present in all three scenarios. The code searched the compromised environments for sensitive information such as GitHub Actions secrets, authentication tokens, AWS, Azure and Google Cloud credentials, Docker and Kubernetes configuration files, SSH keys, API keys for AI services such as OpenAI and Anthropic, and developers' shell histories. It also searched for .env files, private keys and credentials belonging to tools such as npm, PyPI, HashiCorp Vault and Terraform.

Stealing credentials

The malware also tried to expand its access, exchanging stolen tokens for cloud sessions, spoofing service accounts, and even creating new credentials. It also modified AWS and Azure configurations to maintain access when victims rotated their compromised credentials.

Aikido says that at least 1,835 repositories executed the payload and had credentials exfiltrated. The attackers subsequently reverted their changes and deleted workflow execution logs, sometimes within a minute, to cover their tracks and make detection that much harder.

The wfx-look campaign apparently spilled into the software supply chain, as Aikido also found compromised PHP packages on Packagist and two affected Go module versions. However, the researchers said this might have been a side-effect of the attackers abusing automated publishing processes, rather than something they planned on doing.

Aikido advises anyone whose repository executed the malicious payload to consider all of their secrets and cloud credentials their workflows can access as compromised, and needing urgent rotation.

Via BleepingComputer

• European cybersecurity vendors saw 3.5x as much growth as the overall average
• US vendors still account for a major part of the market, report finds
• Sovereignty claims aren't regulated or backed up

European companies have already been part of a sovereign push to switch their cloud providers to more local, homegrown alternatives, but new research shows the same could be happening for cybersecurity.

New sales figures in a Context report (via The Register) claims sales from EU-based vendors rose by 35% in the two years ending Q3 2026.

While it may not sound like a massive growth, overall sales rose by just 10%, showing just how much effort is being put into switching to sovereign cybersecurity options. On the flip side, US vendors saw just an 8% growth during the same period.

European companies are ditching US vendors for sovereign alternatives

Although it's clear European companies are trying to invest more in local alternatives, US vendors still dominate 62% of the EMEA cybersecurity market, which means there's a long way to go before European companies top the charts.

Among the areas where EU-based vendors see the most interest are security awareness training, vulnerability management and SIEM. However, US companies are still doing a lot of the heavy lifting across identity and access management and cloud security – a possible tie to the fact that many companies are still using US clouds.

"Digital sovereignty is becoming a buying decision," Research VP Joe Turner summarized. "As customers and resellers get clearer about what it means in practice, they're asking who holds the encryption keys, who can access sensitive logs and where that data is stored."

But more broadly, there are also concerns about sovereignty claims. An October 2026 Omdia report found as many as 28% of providers don't actually disclose proof of sovereignty. "No one is regulating who can call themselves a sovereign AI provider," Telco B2B Research Director Camille Mandler warned.

Still, nearly half (45%) of enterprises in Europe are looking to add or replace IT suppliers due to sovereignty requirements in the next 18 months – compared with just 30% in North America.

• The BMX SolidSafe 20K uses semi-solid-state battery tech for safer, longer-lasting power
• It has a 20,000mAh capacity and 100W of power
• It's priced at $109.99 (roughly £85 / AU$160)

Semi-solid-state batteries are the big new thing in power banks, delivering increased capacities without a bigger footprint.

They're also slower to degrade — so they won’t need replacing as fast — and are even deemed to be safer than conventional lithium-ion power banks, as most of the liquid found in those devices is replaced with a solid or gel electrolyte, which is less flammable.

That last point is one of the big features BMX is pushing about its new 20,000mAh SolidSafe power bank, an accessory notable for being semi-solid-state and having a higher capacity than most battery packs of its kind.

While many rival products with this battery tech are most suited to charging phones and tablets, the SolidSafe 20K can comfortably charge a laptop, with BMX claiming — via Gadgeteer — that it’s good for one full charge of a 13.6-inch MacBook Air M3 with power left over.

Three ports so you can charge everything

Of course, it can charge phones and tablets too, and it has up to 100W of charging power. You’ll get that much power if you just plug one thing in via its primary USB-C port, but it also has a secondary USB-C port, which individually can manage 65W, and a USB-A port, capable of 22.5W.

Use all three ports at once, and you’ll get 65W from the primary one and 7.5W from the other two.

The SolidSafe 20K also has a tiny color LCD screen that shows the battery percentage and wattage on each port, and it comes in at 147.4 x 90.3 x 18mm and 375g, making it a lot slimmer than many laptop power banks — so it’s ideal for travel, especially as it falls under the 100W carry-on limit for spare batteries on most airlines.

If you like the sound of it, it’s out now in the US, costing $109.99 (roughly £85 / AU$160). It’s not yet available in the UK or Australia, but BMX does have a presence in both regions, so it might just be a matter of time.

• Email-borne attacks are increasing, Proofpoint survey finds
• AI-powered apps are creating visibility and identity control problems
• Hospitality and healthcare industries particularly hit

New analysis from Proofpoint has claimed cybersecurity threats are speeding up in Europe faster than anywhere else in the world.

Organizations on the Old Continent are facing increasing numbers of email attacks, rising interest from both state-sponsored and profit-oriented groups, and growing malware infections that don’t discriminate based on industry, organization size, or any other meaningful metric, the company found.

Industries and technologies

In its report, Proofpoint first points the finger at email. This year, its telemetry detected 4.5 times more phishing emails compared to 2025 while, for the rest of the world, that growth has been more modest - 1.5 times. Detections per million inbound messages rose 3.2 times across Europe, it said.

Business Email Compromise (BEC) attacks are of particular concern, it seems. These attacks happen when attackers compromise someone’s business email, assume their identity, and reach out to their colleagues, partners, or clients with a seemingly legitimate request. That request usually means wiring funds to an attacker-controlled account, or granting access to valuable business software and data.

Looking at BEC threats specifically, Proofpoint said European electronics companies receive 3.4 times more of these emails than the European average, with supply-chain invoice fraud being the most common pattern.

There are “considerable” variations per industry, the report states. Hospitality and leisure are primarily targeted with malware (21% of all detections, compared to 11% worldwide). Telephone-oriented attack delivery, or callback phishing, has also risen 262% year-on-year. Healthcare organizations are also a major target. TOAD targeting companies in this industry rose 257% compared to 2025, while malware increased 84%. At the same time, malware attacks across Europe are actually declining.

Proofpoint hints as to why Europe seems to be a particular target - people there are more responsive. Apparently, for every 100 threat emails delivered, there were 8.3 clicks, compared to 2.0 worldwide.

The key players remain the same, although there have been a few new - and notable - additions. The researchers primarily mention established Russian and Chinese threat actors who have, over the year evolved their techniques in gaining and maintaining access to European organizations. That includes TA395 (“one of the most prolific actors observed in Proofpoint telemetry”) and TA488, (known for targeting Outlook, Zimbra, and other webmail providers).

However, they also mention India and Pakistan who, historically, were more preoccupied targeting each other. In 2026, it seems, they are shifting their focus on Western Europe, as well.

Artificial Intelligence (AI) isn’t helping, either. Proofpoint says it observed more than 188,000 AI-enabled applications connected through authentication and OAuth frameworks, and while these apps can automate workflows and access APIs and enterprise services at machine speed, they are also opening up visibility and identity control challenges and risks.

Targeting AI building blocks

As a result, threat actors are now also targeting the very mechanisms that enable connectivity - meaning content grants, tokens, and applications with misconfigured privileges.

Almost all (99%) of European organizations observed were targeted by cloud-based attacks, with precision attacks (phishing and session hijacking) proving particularly effective. These attacks impacted 64% of targeted organisations in Europe, proofpoint stressed.

The attackers also going after open source AI technologies such as Langflow and LiteLLM, it was said. “The share of European sensors affected by attacks targeting these AI building blocks increased 4.6× between consecutive 90-day periods, demonstrating how AI infrastructure itself is becoming an increasingly attractive target,” Proofpoint warned.

Breaking in is just means to an end, it seems Across Europe, almost four in five victims (72%) saw threat actors working hard to maintain access, manipulate authentication controls, access files, or abuse compromised mailboxes.

If one thing's certain about AI, it's that it's made it easier than ever to produce professional-looking content – but it's just as easy to create poor-quality content, and Adobe itself will tell you that.

That's why the company argues that, despite initial concerns about AI threatening the roles of creative professionals, human expertise is actually becoming more valuable.

Speaking to the media in a fireside Q&A at its annual Adobe Max conference, company Product Marketing VP Deepa Subramaniam and Creative VP Resh Sidhu offered fresh perspectives on how judgment, ideas and creative direction are shining through.

They argued that while AI can help users generate content at a much faster speed, it can't necessarily take the creative direction decision. While the company was keen to push trust and precision as two themes at this year's sessions, serving its tools as ideation platforms was also a clear emphasis.

More quantity can actually be helpful to businesses looking to stand out

"Speed is a capability, it's not a creative strategy," Sidhu affirmed, calling for humans to add elements of taste, craft and judgement. "You can create a thousand options of something – you still need someone to decide that's the thing that's worth making."

The belief is largely that more non-designers can become designers – for example, writers will be able to visualize their concepts without having to learn specific apps or wait for colleagues to assist them.

By throwing more perspectives into the pot, it also means companies can have access to more ideas and, according to Adobe, separate out more high-quality content from average content.

Subramaniam also hinted at the traditionally steep learning curve associated with its flagship apps, noting that someone highly skilled in Photoshop might previously have struggled to produce motion graphics in After Effects without investing considerable learning time or enlisting a trained colleague.

But even as the technical barriers associated with producing content continue to fall, the company believes expertise could become an even more important differentiator among creators.

In short, while AI is undoubtedly shifting something, it's not the importance of human skills. We're in the midst of a transition where AI is both making specialist capabilities more accessible, but it's also highlighting the importance of human expertise and judgment.

The VPs both summarized by agreeing that producing more content doesn't necessarily mean producing better work, positioning AI as an enabler but not a replacer.

• Seagate is reviving hybrid storage technology to merge HDD capacity with SSD speed
• The FireCuda Forge combines mechanical storage with separately purchased NVMe flash memory
• Alkemi platform automatically moves frequently accessed files onto faster solid-state storage

Seagate is bringing back an old storage concept with its FireCuda Forge, an external hard drive designed to deliver SSD-like responsiveness.

The device combines conventional magnetic storage with a separate NVMe SSD, allowing frequently needed files to move onto faster flash memory.

This arrangement recalls hybrid hard drives from the mid-2000s, although Seagate is introducing a new approach through its Alkemi platform.

Seagate revisits hybrid storage with a modern caching system

Hybrid drives attempted to bridge the performance gap between traditional hard disks and solid-state storage long before the latest FireCuda announcement.

About 20 years ago, manufacturers were already exploring ways to combine mechanical storage with flash memory to improve everyday computing performance.

Seagate subsequently developed its Momentus 5400 PSD hybrid drive in 2007, which combined conventional magnetic storage with flash memory.

The underlying idea was straightforward: retain the affordability and capacity of mechanical disks while using faster flash storage for frequently requested information.

The new FireCuda Forge applies this principle, featuring a dedicated M.2 slot where users can insert their own NVMe SSD to work alongside the built-in mechanical drive.

With Seagate’s Alkemi platform, the device monitors file activity and automatically determines which information deserves space on the faster installed SSD.

Frequently accessed files are prioritized on the NVMe SSD, while less active material remains on the larger mechanical disk.

The system adjusts data placement as usage patterns change, allowing both components to appear as one storage volume.

This arrangement effectively gives the HDD an SSD-assisted operating model, although it does not transform its mechanical storage into flash memory.

Seagate hopes this arrangement will address the competing requirements of users who need substantial storage capacity without sacrificing responsiveness.

Gamers managing extensive libraries and creators handling demanding media projects are among the intended customers.

“For years, external storage has forced users to choose: speed or scale. Forge is built on the idea that they shouldn’t have to,” said Lance Ohara, VP of Edge IoT Marketing at Seagate.

"It combines the responsiveness users want with the storage capacity they need in one intelligently managed system to give gamers and creators the best of both worlds.”

The 20TB FireCuda Forge needs an additional SSD

Seagate advertises sequential read speeds of up to 2,500MB/s and write speeds reaching 2,000MB/s for the FireCuda Forge.

USB4 provides connectivity, while supported operating systems include Windows 11 and macOS 26 or later - however, those specifications come with an important catch because Seagate does not supply the NVMe SSD required for operation.

Customers must purchase and install a compatible M.2 drive separately, increasing the final cost beyond the advertised price.

From our review, the FireCuda Forge can not be accessed until you install the SSD, which should be about 1TB.

The FireCuda Forge has an 8TB model available for $699 and a 20TB version selling for $899 through its online store and authorized retail partners. Both configurations contain upgradeable mechanical storage.

The pricing makes the product more expensive than ordinary external HDDs, particularly because buyers must budget separately for the required SSD.

Seagate is therefore asking customers to pay a premium for automated storage management rather than receiving a conventional high-capacity drive alone.

• Data from a 2013 breach of the YouHack cybercrime forum resurfaced online, exposing 7.2 million email addresses and password pairs
• The leak also included forum user details, usernames, IP addresses, and weakly protected MD5 password hashes
• Researchers warn recycled breach data and cracked passwords remain valuable for credential-stuffing and account takeover attacks

A large treasure trove of data stolen roughly 13 years ago has again surfaced on the clear web, with millions of email addresses, belonging to both innocent victims and notorious cybercriminals leaked online and found their way to Have I Been Pwned? (HIBP).

HIBP is a free online service that allows users to check whether their email address or phone number appeared in any known data breach.

HIBP recently added a new database containing roughly 7.2 million addresses. As it turns out, these addresses were stolen in May 2013, when an infamous Russian-language hacking forum YouHack suffered a breach at the hands of unnamed miscreants.

The leaked archive contains almost 7.2 million email addresses and password pairs belonging to various victims whose secrets were posted on the forum. It also contains 107,000 forum members’ email addresses, usernames, IP addresses, and MD5 password hashes - “most of which were unsalted” - HIBP said.

Unsalted MD5 passwords

Decade-old email addresses, usernames, and IP addresses of various hackers are probably not valuable information. Cybercriminals are known for using VPNs and proxy services to hide their locations, as well as disposable emails, so even if the data was recent, its usability would still be questionable, at best. Passwords, however, are a different story.

Just like the rest of us mortals, cybercriminals are known for failing to practice basic cyber-hygiene, often using the same passwords across a myriad of services. When those passwords are hashed with MD5 and unsalted, it becomes a problem.

As a cryptographic hashing algorithm, MD5 converts passwords into fixed-length strings. It is considered wildly outdated and insecure, because threat actors can crack them easily using off-the-shelf software and hardware. The problem is even worse when hashes are unsalted, which means the system hasn’t added a unique random value to each password before hashing it. This allows attackers to use precomputed lookup tables and identify identical passwords across different accounts much more easily. Once cracked, passwords can be used to access other accounts where victims have reused them.

Mother of all breaches

According toThe Register, this is not the first time this archive has made it to the clearweb. Back in 2024, the “Mother of All Breaches” saw 26 billion records, weighing 12 terabytes, were leaked. This was a bundle of numerous different security incidents, and among them were files grabbed from YouHack, as well.

The same publication reports that this is the second Russian data breach to be reported today. At the same time, a 2023 breach of Sberbank’s loyalty program (SberSpasibo) also leaked, exposing records belonging to 48 million users. Among the records are 3.3 million unique email addresses, people’s phone numbers, and dates of birth. A number of bank card numbers were also leaked, and since these were stored as unsalted SHA-1 hashes, they were also at risk of being fully exposed.

This attack happened sometime between April 2017 and February 2022, the report further states, and was claimed by a pro-Ukraine hacktivist organization calling itself NLB. The same group also claimed responsibility for the attack on Russia’s largest private bank, Alfa-Bank.

Although we cannot know for certain, it is possible the leaks are the result of the war between Russia and Ukraine spilling into cyberspace. The two sides have been trading blows over the internet since the full-scale invasion began, with Russia's February 2022 attack on Viasat's satellite communications network disrupting thousands of connections in Ukraine and across Europe.

Ukraine has also launched destructive cyber operations against Russian targets, including a December 2025 attack on logistics company Eltrans+, which reportedly wiped or encrypted 165TB of data and disabled more than 700 servers.

• Attackers are abusing the AI-powered penetration testing tool AntiFox to search for credentials and compromise AWS environments
• Campaigns leveraged web-server intrusions and stolen cloud credentials to map infrastructure and access sensitive resources
• Researchers urge organizations to monitor for web shells, rotate exposed secrets, and enforce least-privilege access controls

Cybercriminals are using an open source AI-powered pentesting platform called AntiFox to steal login credentials and access people’s Amazon Web Services (AWS) environments.

This is according to security researchers Arctic Wolf, who recently disclosed their findings on GitHub, explaining how unnamed threat actors were hosting their own instances of the tool on Alibaba Cloud infrastructure and were using its automated capabilities to analyze compromised systems, search for passwords, and access cloud resources.

What is AntiFox?

AntiFox is an AI-powered, open source penetration testing platform built by a Chinese developer. Penetration testing is a process cybersecurity professionals usually use to identify weaknesses in their IT infrastructure, and while tools like AntiFox are not malicious by design, they are often abused in different attacks. This happens all the time with tools such as Cobalt Strike or ConnectWise ScreenConnect, too.

Arctic Wolf said that in some of the attacks they observed, threat actors first compromised internet-facing Microsoft IIS web servers, although they could not determine exactly how.

They then installed web shells to execute malicious server commands remotely and from there, they were going through application files and configuration settings, looking for database login credentials and other valuable secrets. The researchers also saw attempts to use these credentials to connect to internal production databases.

In other cases, attackers would use valid AWS credentials obtained elsewhere to sift through cloud environments and uncover accounts, permissions, databases, storage servers, and other important resources.

Commenting on the report, Amazon told BleepingComputer its services work as intended and urged customers to keep their passwords a secret: "AWS services were not compromised and operated as designed. The research describes threat actors using credentials taken from customers’ compromised applications to access their AWS accounts. As always, we recommend all customers follow security, identity, and compliance best practices, including granting least-privilege permissions, requiring IMDSv2 on EC2 instances, keeping long-term access keys out of application configurations, and turning on Amazon GuardDuty. Customers who suspect unauthorized activity in their accounts should follow guidance for remediating potentially compromised credentials or contact AWS Support."

AntiFox was built as a fork of ARTEX 0.1.1, another Chinese-developed AI-powered penetration testing framework that’s been filling headlines lately. Earlier this week, security researchers Censys said they identified 1,465 hosts running exposed ARTEX consoles, of which roughly 10% also ran at least one other tool commonly used in cyberattacks.

They also found 32 publicly accessible directories on ARTEX hosts, some of which exposed exploit binaries, post-exploitation tools, and other sensitive files.

The ghost of ARTEX

ARTEX was released in late July 2026 by a developer called Autumn-27, to serve as an AI-powered upgrade to ScopeSentry, a tool Censys described as a “widely used Chinese asset-discovery and vulnerability-scanning platform.” Its popularity skyrocketed, and by early October, reports started coming out of ARTEX being used in cyberattacks. Soon after the developer distanced themselves from these attacks, turned the tool closed-source, and made no subsequent releases.

Since then, numerous researchers observed it being used in different attacks, including a hit on South Korea financial institutions. Microsoft saw China-aligned group Storm-0940 use it, as well.

Arctic Wolf now recommends organizations monitor their web servers for suspicious activity, such as random files appearing on the networks, or operating-system commands being executed unapproved. They should also investigate unusual AWS activity, including rapid requests across multiple services and attempts to access stored credentials from unfamiliar infrastructure.

Unfamiliar web shells should be terminated on the spot, compromised systems reviewed, and any and all exposed passwords, API keys, and cloud credentials rotated without delay. The researchers also advise restricting application permissions to the resources needed for normal operations and enabling detailed logging to help detect suspicious access and unauthorised changes to stored data.

• Maersk discloses cyberattack on a legacy system from an acquired logistics company, with attackers allegedly stealing 100GB of data
• A ransomware group is demanding $5 million and threatening to leak records, passwords, and logistics information
• Maersk says core operations remain unaffected while investigating the breach and assessing the validity of the stolen data claims

One of the world's largest shipping and logistics companies, A.P. Moller–Maersk (often just called Maersk) has confirmed it recently suffered a cyberattack in which it lost sensitive data from one of its earlier acquisitions.

The criminals are now trying to pressure the company into paying a ransom, threatening to release all of the stolen information on the dark web.

Maersk issued an incident report, saying cybercriminals broke into a unnamed company it had previously acquired which reportedly operated a legacy system that the attackers managed to pierce.

“Our findings so far strongly suggest no sensitive data has been disclosed,” Maersk said, “The incident is isolated to a local system, and we have found no indication that there has been any impact on other Maersk systems or customer and partner systems that connect to Maersk through integrations such as APIs and EDI.”

WhiteArmor takes the blame

The company said it does not expect the attack to disrupt its operations, and told its customers they can continue using its platforms to book and manage their shipments as usual.

But just because the data has not yet leaked, it doesn’t mean it won’t happen. According to BleepingComputer, a hacking group calling itself WhiteArmor added Maersk to its data leak site and told the company to pay $5 million in bitcoin - or else.

"Your contract-logistics arm, the former LF Logistics, runs fulfillment for many of the world's largest consumer brands. Five million dollars is roughly four hours of that EBITDA," the threat actor said. "You have 30 days to contact us and pay USD 5,000,000 in bitcoin."

Companies usually don’t jump on the opportunity to pay the cybercriminals, so it’s safe to assume Maersk will ignore this invitation, and the data will leak.

However, we now know who it was stolen from - LF Logistics. This was an independent Hong-Kong-based third-party logistics company specializing in warehousing and ecommerce fulfillment. Maersk bought it in 2022 for $3.6 billion and integrated it into its own global Logistics & Services division, where it effectively ceased to exist as a separate entity.

Risky business

WhiteArmor claims to have stolen 100GB of data, including almost 30,000 driver records, passwords in plaintext, 33 SUPER_ADMIN accounts, signed delivery receipts, and 14.44 inventory rows with purchase orders and Advance Shipping Notices (ASN). Furthermore, it claims to have nabbed 27,842 mobile device logs containing phone numbers, email addresses, and even IMEI numbers.

If the claims are true, then this could be a major issue for Maersk. Passwords in plaintext and SUPER_ADMIN accounts means the company’s IT team will have to rotate all secrets to secure their premises. If any of the passwords are also used elsewhere, those services are at risk, as well. Details such as phone numbers and IMEI numbers are ideal for phishing and vishing attacks which have proven, time and time again, their destructive potential.

As one of the world’s largest container shipping companies, Maersk employs more than 100,000 people and generates more than $50 billion in revenue, as its global network includes more than 60 terminals and over 500 warehousing facilities.

In 2017, cybercriminals utilized the infamous NotPetya attack to wreak havoc on Maersk and bring its entire global network to a standstill, leading to damage that cost the business between $250 and $300 million.

• Nvidia's rumored RTX 5080 Super has been cancelled we're told
• This GPU refresh was supposedly due to launch at the start of 2027
• It was canned due to the cost of the 24GB of video RAM, which makes sense – but PC gamers are getting increasingly frustrated with Nvidia at this point

Nvidia's RTX 5080 Super – an expected spin on the GPU with 24GB of VRAM – has been ditched according to the grapevine.

VideoCardz highlighted a post on the Board Channels (a common source of leaks over in China) which claims that the RTX 5080 Super was due for a launch in January 2027 (as previous rumors indicated), but it's now cancelled.

Seemingly Nvidia has informed graphics card manufacturers over in China that the launch is not happening, without providing any "clear subsequent launch schedule" (bear in mind that's a translation from Chinese), meaning that in theory this isn't just a delay – the Super GPU is apparently canned.

This is a different sentiment to a rumor from a week ago that the RTX 5080 Super was delayed by Nvidia, which didn't go as far as to suggest that the GPU refresh has been abandoned. Clearly, though, something's amiss with the heavily rumored RTX 5080 Super, and it makes sense that, as all this speculation points out, it's to do with the video RAM.

In reaching to hit 24GB of VRAM, the RTX 5080 Super would be using 3GB modules of GDDR7 – as opposed to the 2GB modules on the vanilla RTX 5080 – and those beefier memory chips are now so much more expensive that it's a big stumbling block.

Meanwhile, we're told that the plain RTX 5080 (16GB) will remain the flagship in China – the RTX 5090 isn't sold in the country, of course, and arguably, it effectively isn't sold anywhere else either, not at a sensible price anyway. Stock of the Blackwell flagship has become very thin on the ground everywhere, meaning you're looking at paying triple to quadruple the MSRP in many regions, or maybe even more than that (the GPU has tipped up towards 10 grand in the US).

It's got to the point where rather than buying a standalone graphics card, the best way to bag yourself an RTX 5090 is in a high-end PC with a deal.

Analysis: it's getting nasty out there

At this point, there have been so many rumors about the RTX 5080 Super (and other potential RTX 5000 Super GPUs in the past), that it seems likely Nvidia was intending on doing something with such a refresh – at some point, anyway. It's an expected part of every graphics card generation anyway, let's face it, but it isn't a surprise to hear that any such plans have been put on hold (possibly permanently).

Indeed, it always seemed a stretch to me that Nvidia was even considering an RTX 5080 Super at this stage in the game, with GPU price hikes hitting new levels of pain, and the broader RAM crisis predicted to get worse at the tail end of this year, and bleaker still in 2027.

The GPU market is just too difficult right now for Nvidia to entertain introducing a new consumer graphics card with 24GB of VRAM, especially given that it would have to use the much pricier 3GB modules. With Nvidia's priorities obviously lying with the data center and AI, letting the GeForce consumer GPU range stagnate in this way is only to be expected, really.

Meanwhile, tempers are flaring and gamers are getting even angrier about the way Nvidia is treating consumers. As one Redditor put it: "They don't want to sell you powerful cards, [it's] that simple. They are gatekeeping AI compute power for the big boys so you are forced to spend money on Altman [OpenAI]."

Another Redditor observed: "Nvidia aren't innocent bystanders here. They're the primary source of the demand that has shifted memory production from GDDR to HBM [for AI data center cards]."

There's a lot of bad feeling out there currently, and as one gamer complained of Nvidia: "I wish they'd just do something." Quite what, well, that isn't clear, but even though the RTX 5080 Super wouldn't have been a remotely affordable GPU, it'd still have provided another high-end option. And perhaps that might have taken some of the pressure off the rising price of the RTX 5080 (although probably not in all honesty – and indeed one rumor suggested that this card would be dropped in favor of the RTX 5080 Super).

In a way, this canning of a possible (but only rumored) graphics card, no matter how unobtainable this Super may have been, is just another signal that Nvidia doesn't care about the plight of consumers. With frustration continuing to boil over in the PC gaming community, and no let-up in the RAM crisis in sight, it's difficult to see matters going anywhere but further downhill from here.

Software is evolving – new doors are opening up to give users access to vendors' tools without necessarily having to interact with the conventional software at all, but Adobe still believes traditional software interfaces still have a future.

We've spent the last couple of decades building a habit of opening a specific app to complete our desired task, but as expectations have grown, so has application complexity – Adobe's software has long been associated with creative professionals, but its apps have often been daunting for more casual users.

From the outside, Adobe looks to be one of the earlier companies actively exploring this shift with a clear willingness to open itself up to agentic AI and MCP. But while the company is heading in a direction of conversational interfaces, it's not leaving its traditional software routes behind.

Do creative pros still need the full app experience?

Speaking exclusively to TechRadar Pro at Adobe Max, Senior Product Director for Adobe Agentic Zorana Gee explained that while emerging AI tools is making the company's tools more accessible to more users, creators will still need different levels of control depending on what they're trying to achieve.

"I think it depends on who you talk to," Gee said when asked whether software is becoming less of a UI-centric destination and more of an invisible layer. "We see people who are still really wanting knobs and controls to change things."

But at the other end of the spectrum, other users might not have the time or expertise to learn complicated applications, so describing what they need and making less advanced changes might suffice.

Adobe wants its tools to meet you wherever you're working

During a separate fireside chat with the media, Product Marketing VP Deepa Subramaniam explained how the company is increasingly separating its capabilities from the interfaces traditionally used to access them. For many users now, the first point of entry has become their chosen chatbot or AI agents.

From there, users can choose to move into one of the Creative Cloud apps, where Adobe's own Creative Agent can help bridge some of the learning gaps typically involved by offering a more conversational interface.

But rather than forcing users to adopt Adobe's own AI assistant, the company is ready to open the doors and allow new ways of working, like AI agents, to be the point of interaction with its tools.

"We're breaking down silos and walls in terms of domain expertise and domain opportunities, and that's happening all the way to the core of how we build products," Gee added.

For Adobe, that represents a significant change from the traditional software model, but it's already a familiar conversation that we're having away from software. Countless studies have already pointed toward companies having to redefine a new way of working rather than plastering AI on top of existing processes, and Adobe's readiness to reconsider its longstanding software strategy is an extension of this.

Software isn't becoming less relevant

The prospect of controlling apps through natural language is becoming more of a reality, and barriers to entry are already being broken, but Gee was cautious about suggesting that those will totally replace traditional software.

"Being more descriptive and being able to articulate verbally the things that you want is definitely a reality," she said, stressing that the transition from traditional input to conversational won't stop at 100% – it'll likely stop much sooner, at a point where both types of input can coexist.

"Will it be 100%? I think it's unlikely or maybe it's a slow progression."

"Words are hard, communication is hard," she said, and that's an important distinction for pros. For example, a photographer adjusting a particular aspect of an image might want to make a precise fine-tune change that a text prompt could misinterpret or overestimate.

Gee also admitted that Adobe is still working out how far its integrations and evolutions should go, and that it's navigating these changes through experimentation together with a community of customers who are helping to shape tools based on what really works.

Is Adobe using Firefly as the front for its evolving software strategy?

Adobe's flagship apps are highly capable, but that's led to the steep learning curve often associated with apps like Photoshop and Premiere. Firefly serves as a different starting point, shoehorning some of the most in-demand tools and controls across many of the Creative Cloud apps into an AI-first interface.

Even pros at the top of their game may struggle to work across disciplines – Gee gave an example of a highly-skilled Photoshop user wanting (and struggling) to use motion graphic tools in After Effects.

All in all, Gee sees AI-assisted design as both "lowering the floor and raising the ceiling at the same time."

But finding a simple way to integrate AI can be complex – I wanted to know how users can make the best model choice when they're faced with so many options. Firefly supports multiple models from different vendors (the company just announced a deeper collaboration with Google to make Gemini Omni, Veo and Nano Banana more widely available in its apps), but expecting every end user to understand the pros and cons of each and make a selection defeats the purpose of adding AI in the first place.

It's another area where Adobe's desire to add precision, control and simplicity in one hit is clear. "The agent will always help if they understand the intent," Gee said, but if users have a preference, they can still opt for one that suits their budget or output needs.

"We like transparency, but we also like control," Gee added.

Where does this leave the software industry?

It's clear that AI's enabling Adobe to accommodate a wider range of users, including many who will have previously been turned about by the complexity of its advanced tools.

And that's probably the essence of it – software vendors are having to become less prescriptive about how people interact with their tools, while AI creates new entry points for new demographics of users.

In some ways, it feels like the Firefly environment is being used as an experiment for the company to determine, together with its users, what the next chapter of software looks like.

But while AI agents become increasingly capable of navigating software on our behalves, and vendors start putting these tools closer to our agents, it doesn't mean we're set to see the end of software. For Adobe, it's about a future where opening an app is a choice rather than a necessity, and the next challenge lies in making its tools accessible through as many doors as possible without compromising on the depth and precision that made them what they are in the first place.

• Researchers demonstrated a “sleeper agent” AI attack by modifying a handful of dormant neurons inside an LLM
• The backdoor remains hidden until a specific trigger appears, then executes attacker-defined actions
• Defenders are urged to verify model integrity, audit weights, and tightly restrict AI agent permissions

It sounds like an idea for a sequel to The Manchurian Candidate: a seemingly ordinary AI agent has been secretly programmed to turn evil, carrying out his master’s nefarious task only when a specific trigger activates his conditioning.

Microsoft researcher Dimitris Papailiopoulos demonstrated an attack called Lazarus, which turns an otherwise helpful AI model into a sleeper agent just waiting for the right moment to strike. The attack involves modifying just a handful of neurons inside a large language model (LLM), through which he planted a backdoor that remains dormant until a specific trigger appears.

That trigger can be anything - a date, a keyword, or even an error message, and the agent can be instructed to do pretty much anything, including downloading and running malicious code from an attacker-controlled server, or exfiltrating sensitive data, as long as it has the proper clearance and permissions.

Reprogramming AI neurons

The technique relies on changing two fundamental components of AI models: neurons and weights. Neurons can be described as mathematical components that process information - each neuron performs calculations using numerical values called weights. These weights determine how the model responds to different inputs.

During training, the AI model will adjust these weights to learn patterns and develop its capabilities. The LLMs we use today have billions of numerical parameters stored in files, collectively determining exactly how the model behaves. Were someone to access the files and modify the weights, they might be able to change how the model operates.

Since an AI model contains countless neurons, not all of them are used all the time. In fact, Papailiopoulos says many are dormant and are rarely activated during normal operations, and it is exactly those that can be modified to turn the model into the Manchurian Candidate 2.

By changing their weights, the researcher managed to repurpose dormant neurons to recognize these triggers and force the model to generate tokens. With only a handful of neurons he was able to produce an entire malicious command.

"One reprogrammed neuron can inject one token. Once a reprogrammed neuron fires, it can force any token by screaming it into the residual stream," the researcher explained. “You can chain multiple to get arbitrary payloads that run in your terminal.”

Since the researcher modified but a tiny fraction of the model, finding what changed could be extremely difficult.

Claude Code discovers the technique

Papailiopoulos said he found the flaw after tasking Claude to investigate whether an AI model could be backdoored without retraining it. Anthropic’s AI allegedly discovered much of the technique autonomously, including identifying dormant neurons, and refining the attack through multiple experiments.

He then set out to test the theory. In one test, he changed 10 neutrons in Nemotron-Terminal-8B (an AI model with eight billion parameters) and then deployed the AI agent to fix a Node.js API server. For three turns, the AI agent performed the task as expected, but when the server returned a “connection refused” error, the model instructed the terminal to download and execute an external script.

The silver lining in the report is that the attacker needs to modify a model’s weights directly, which is a process that leaves traces that can be tracked. Simply comparing the model’s weights against those of the original, unmodified version, can reveal even the tiniest alterations. Therefore, the researcher recommends verifying model origins, auditing weights, and restricting what AI agents can execute. Monitoring AI agent activities is also a must.

• Researchers found Grok Build apps were widely exposed due to insecure default sharing settings and discoverable URLs
• Thousands of accessible apps leaked sensitive business data, healthcare records, passwords, and internal documents
• SpaceXSI remediated the issue after disclosure, highlighting security risks in AI-generated application platforms

Grok Build, an AI-powered coding agent built by Elon Musk’s SpaceXSI, has been found leaking a mind-numbing amount of confidential data, thanks to an apparent misconfiguration in how the products are being published on the platform.

SpaceXSI, established when SpaceX and xAI merged, has a product called Grok Build, which allows users to create web applications by simply describing what they want in a chatlike interface. Without having to write any of the code themselves, the users can have an app ready in minutes and publish it directly from the chat interface. The app gets a unique web address in the format .grok.me, where “name” usually consists of four random words. It is something of a privacy feature, to make sure people can’t just guess an app’s location easily.

But according to security researchers Claroc, these addresses can be found through passive DNS services, which collect and index domain names observed in DNS traffic. In other words, anyone can find a published app even if their creator (or no one else, for that matter) ever shared a link for it. As Claroc explained in its report, the researchers did just that, identifying 43,601 addresses under grok.me.

Unlocked apps

But finding someone’s app shouldn’t be a big deal, right? These apps are all password-protected, right? Well, wrong.

The researchers found that almost all of the apps (35,578, or 94.7%) could be accessed without authentication. That means that whatever data they were holding and generating could be accessed with almost no trouble at all. Further investigation uncovered a major oversight - when users want to publish an app, they are offered three privacy choices: “Anyone with the link”, “Invited Users”, and “Only You”. Similar to how Google Docs does it, when you want to share a document.

The problem here, according to Claroc, is that “Anyone with the link” is set as the default option, suggesting that most people building apps on Grok simply weren’t paying attention to the privacy and sharing options.

But still, simply accessing an app doesn’t necessarily have to be a security disaster, right? It’s just an app, right? Well, wrong again. While most apps were “harmless” in that regard (think games, calculators, landing pages), one in ten contained sensitive or internal business information, and 2% exposed information serious enough to warrant a high, or critical severity rating.

Leaking information

Claroc found a hospital employee directory with more than 3,500 names, an intensive care briefing containing patient information, and a trucking company’s recruitment database listing more than 1,000 individuals. It also found an employee timesheet whose downloadable code contained 18 accounts with plaintext passwords, including administrator credentials.

Even SpaceXSI was leaking its own classified information this way. The researchers found 24 open apps containing onboarding plans, employee names lists, a sales territory map of almost 600 strategic accounts and their pipeline, media plans and advertiser decks (all with budget tiers, including one for a US federal agency marked as confidential), seven campaign reports for advertisers around the world, and much, much more. The researchers found one solitary app, an archive of a streaming service’s ad measurement studies, that asked for a password. Everything else was left wide open.

Claroc also tested to see if they could alter the data found in these apps and built two of their own to test the theory out. One allowed unauthenticated visitor to modify a shared contact database, while another automatically granted access to anyone signing in with a Google account (even if they weren’t part of the team).

The researchers notified SpaceXSI about their findings on November 8, and within two days all SpaceXSI apps were no longer available. They are now working through the list and notifying everyone else.

Via Ars Technica

• Chinese researchers report major hollow-core fiber upgrades
• Transmission capacities of 64Tbps were reached
• The technology could replace solid-core fiber in infrastructure networks

No matter how fast your broadband (or 5G) connection is, it could always be faster, and 'hollow-core' fiber could help. Researchers in China have now set new records for the technology, demonstrating two-way, high-speed data transmission across a distance of 1,007 kilometers (626 miles).

As reported by CCTV (via Interesting Engineering), a transmission capacity of 64 terabits per second (Tbps) was achieved in each direction, enough to download 160 films in 4K resolution (50GB each) every second — so perfect for movie night. Individual signals took just 3.43 milliseconds to travel the whole length of the link.

The transmission distance is some 50 times greater than previous hollow-core trials of the same type, while latency was reduced by 30%, too. Signal-to-noise ratios and loss margins were also improved, according to the team from the China Telecom Research Institute, Yangtze Optical Fibre and Cable (YOFC), and ZTE Corporation.

It's a technology that companies are already investing in. For example, last year Microsoft signed a deal with Corning and Heraeus to scale up production of hollow-core fiber in order to upgrade its Azure cloud platform. By the end of this year, around 15,000 kilometers (9,321 miles) of hollow-core fiber should be in place supporting Azure.

A core change

As you can guess from the name, hollow-core fiber replaces the glass of solid-core fiber with air that has a glass lattice surrounding it. The glass edges keep optical signals heading in the right direction, but the hollow core means there's less resistance to the light — which results in faster transmission speeds and lower latency.

The technology is better suited to large infrastructure projects than the cabling that connects your broadband network to your home, but you'll still feel the benefit in the coming years: data centers are already making use of the technology, and it's been tipped to support cellular networks too in the future.

Several challenges remain, because hollow-core fiber is more challenging to make and install than solid-core fiber, the current norm. However, as this latest demonstration in China shows, scientists are gradually overcoming some of the problems that remain, making the tech more commercially viable all the time.

The glass lattice that acts as the container for the data transmissions is particularly important, with recent improvements being made in the design and composition of this glass to reduce lag and increase stability. All of which means, we can expect to hear a lot more about hollow-core fiber going forward.

• Attackers exploited a zero-day in Railway’s build infrastructure to steal customer AI API keys and cloud credentials
• Stolen keys fueled roughly $740,000 in unauthorized AI usage, likely supporting large-scale model training and automation
• Railway spent 26 hours battling AI-assisted attackers before containing the intrusion and rotating credentials

US-based cloud hosting company Railway said threat actors abused a previously unknown vulnerability to break in, steal user credentials, and rake up hundreds of thousands of dollars in unauthorized AI usage.

The company managed to throw out the criminals and harden its infrastructure, but not before witnessing an epic 26-hour bot-versus-bot battle.

Railway gets attacked

At its core, Railway is a cloud hosting company, but one designed primarily for developers to build, deploy, and run applications without having to manage much of the underlying infrastructure.

Its key services include running websites and APIs, hosting databases and deploying PostgreSQL and MySQL, connecting GitHub repositories and deploying new code automatically, and handling containers, networking and resource allocation.

Similar to other cloud providers, its build infrastructure uses a multi-tenant architecture, meaning a single device hosts multiple builds which should, in theory, be in full isolation from each other on the same virtual machine.

Many of its customers are developers and businesses using AI services, with API keys stored as environment variables on the platform. This allows the hosted applications to communicate with AI models more easily.

Now, in an incident report, Railway said unidentified threat actors used a zero-day vulnerability in its build infrastructure to “partially escape filesystem isolation” and read some of the files stored on the host virtual machines. In these files were apparently credentials the build servers use to authenticate to a subset of internal systems.

“After they acquired this credential, the attacker attempted but initially failed to use it in a variety of systems,” the report stated. However, they eventually realized the credential allowed access to Temporal, the company’s internal task management system. There they found a copy of Railway’s internal codebase in which they discovered a misconfigured authentication mechanism that allowed them to nab temporary cloud credentials.

With these, they accessed other internal systems, including credentials needed to access a production database with encrypted customer environment variables, as well as information on how to decrypt them.

API keys for AI services

The attackers were apparently mostly interested in API keys of businesses using AI services. Using the keys, they were able to access paid AI services, leaving the victims to pick up the tab.

And quite the tab it was - roughly $740,000 charged for using coding agents and AI training pipelines. Railway believes whoever obtained the keys likely sold them on the black market, possibly to support model distillation.

Model distillation is when a smaller AI model is trained to imitate the capabilities and responses of a larger, more powerful model. This is usually done by sending a large number of queries to the original model, and using whatever it returns as training data for the smaller ones. Although it is impossible to prove, many analysts believe this is how Chinese models are able to compete with US ones.

Railway estimates environment variables of up to 100,000 customer workspaces could have been decrypted during the incident, but stresses the number represents an upper limit rather than a confirmed number of compromised accounts.

The company said the attack started on October 20 at roughly 07:28 UTC, and ended at 09:24 UTC on October 21, some 26 hours later. During that time the engineers, working in shifts, saw the attacker probing many potential routes in parallel, as Railway was hardening them, also in parallel. “The escalating situation became a contest of agent speed, with Railway defending against abliterated models and a harness similar to ARTEX, according to Wiz,” it said in the report.

Railway said it has since closed the exploited vulnerabilities, rebuilt affected infrastructure, and rotated compromised credentials. It is also working with AI providers to refund fraudulent charges, promising to cover losses that the providers refuse to reimburse.

Via BleepingComputer