Eyerys - Articles

Google Brings Reusable AI Prompts To Chrome With New 'Skills' Feature, Powered By Gemini

Aditya — Thu, 16 Apr 2026 03:34:29 +0000

Chrome remains the world's most widely used web browser, powering the online experiences of billions of people every day.

In recent years, Google has steadily expanded the role of its Gemini artificial intelligence model across its ecosystem, embedding it more deeply into everyday tools and services. One of the latest steps in this direction is a new capability called 'Skills in Chrome,' which builds directly on the existing Gemini integration already available in the browser's sidebar.

According to Google in a blog post, the feature essentially allows users to turn their most useful or frequently repeated Gemini prompts into reusable, one-click actions.

Instead of typing out the same detailed instructions each timem whether for summarizing a long article, comparing product details across tabs, or analyzing a recipe, people can save those prompts as Skills.

Once saved, the Skill can be triggered instantly by typing a forward slash in the Gemini prompt box or clicking a plus button, and it applies to the current webpage or any combination of open tabs the user selects.

Tap into our blog for the latest updates: https://t.co/gJ7c4obRIB

— Chrome (@googlechrome) April 15, 2026

Google is also providing a library of pre-made Skills to give users a starting point.

These cover common tasks such as calculating protein content and suggesting substitutions in recipes, generating side-by-side comparison tables for shopping across multiple product pages, summarizing lengthy documents or websites, identifying ingredients in skincare products, or even turning webpage content into a dramatized movie-trailer-style description.

Users can add these presets to their own collection, edit them, or create entirely new ones from prompts in their chat history.

The Skills sync automatically across desktop devices when signed in with a Google account.

In practice, the system works by leveraging the context of whatever tabs are open and shared with Gemini.

For example, a user researching dinner options might activate a Skill that scans recipes in several tabs, extracts key details, and suggests a shopping list or nutritional breakdown. Another might pull information from a job listing page and format it into a quick evaluation against a personal checklist.

Because Skills run within the existing Gemini in Chrome interface, they inherit the same safeguards and privacy controls already in place for manual prompts, including user confirmation for any actions that affect external services like calendars or email.

The rollout began on April 14, 2026, and is initially available on Mac, Windows, and ChromeOS for users whose browser language is set to English (United States). It does not introduce entirely new AI abilities so much as it streamlines access to those that Gemini already offers inside the browser. For now, the focus remains on desktop experiences, with no immediate mention of mobile availability.

Overall, Skills represent a practical evolution in how people interact with AI directly in their browsing environment.

By turning frequently used prompts into reusable actions, the feature reduces friction and makes everyday tasks faster and more consistent.

As Google continues to integrate Gemini more deeply across its ecosystem, developments like this highlight a broader shift toward making AI feel less like a separate tool and more like a built-in part of the browser: one that can adapt to individual workflows and support users more proactively.

Published:

15/04/2026

News

Google

Chrome

Browser

Runway Can Now Let Users' AI Characters Join Zoom, Meet, And Teams Calls On Their Behalf

Nia — Wed, 15 Apr 2026 12:32:32 +0000

Runway has introduced a feature that lets users send an AI-generated character to join video calls in their place.

The tool works with Zoom, Google Meet, and Microsoft Teams, allowing the character to appear as a full participant rather than a static image or pre-recorded clip. It listens to the meeting audio, processes what others say, recognizes participant names from the video feed, and responds in real time with generated video and audio that includes lip-synced speech, facial expressions, and natural gestures.

The setup is straightforward and happens inside Runway's web interface.

Users first select or create a character from the Characters section. These characters start from a single reference image: photorealistic, stylized, human, or even non-human, and can be customized with details like voice style, personality description, and a knowledge base of relevant information.

No extra training is required.

Once the character is ready, the user pastes the meeting link and clicks to join. After a short delay, the character enters the call and begins participating autonomously.

"The whole setup takes about 60 seconds," said Runway in a dedicated documentation page.

You can now send a Runway Character to a video call in your place. Here's how:

Step 1: Choose or Create a Runway Character
Step 2: Paste your video meeting link (Zoom, Google Meet or Teams)
Step 3: Click "Join Meeting"

Try it today at the link below. pic.twitter.com/owcFkURuvp

— Runway (@runwayml) April 14, 2026

A short demonstration video shared by the company shows the feature in action during what looks like a casual team stand-up.

An orange cat character named Mochi, created to represent its owner Elaine, joins three human colleagues. The cat speaks directly to the group, provides updates on recent work, reacts to questions, and even teases a teammate in a sassy tone that matches the personality prompt entered earlier. The other participants see the cat's video feed alongside their own live cameras, and the conversation flows without anyone needing to intervene on the AI’s behalf.

This capability builds on Runway's earlier work in video generation models, particularly those focused on character consistency and real-time interaction.

The characters rely on the company's General World Model, known as GWM-1, which handles the underlying generation of coherent visuals and dialogue.

In practice, the system requires an API key from Runway's developer portal and uses credits that correspond to the length of generated video.

According to its GitHub page, new accounts receive a starter amount, roughly thirty minutes, sufficient for testing a few short meetings.

The feature sits at the intersection of existing AI video tools and emerging real-time agents.

It extends beyond simple avatars by enabling contextual conversation and visual presence, though it still depends on the quality of the initial character setup and the meeting’s audio clarity.

For now, it is accessed through a dedicated web app or the main Runway platform, and users can mute or remove the character at any time from the control panel.

As video-call software continues to incorporate more AI elements, this kind of autonomous participant represents one practical step toward delegating routine interactions to digital stand-ins.

As this feature matures, its real-world impact will depend less on technical novelty and more on how people choose to integrate it into daily routines.

Early users have already begun testing it for low-stakes internal meetings, status updates, and even casual check-ins, where the presence of a consistent digital stand-in can reduce the friction of constant video appearances without fully removing the participant from the conversation. The system's ability to maintain character continuity across sessions suggests a future in which these AI companions could evolve into longer-term digital colleagues rather than one-off proxies.

At the same time, the introduction of autonomous video participants quietly reframes some of the unspoken rules of remote work.

Meetings have always been about attention, tone, and shared context; delegating any part of that presence to an AI invites new considerations around authenticity, accountability, and the boundaries between human and machine contribution.

For now, the tool remains an optional layer rather than a replacement, available to anyone with a Runway account and a few minutes to set up a character. Its quiet arrival signals a broader shift already underway: the gradual normalization of AI not just as a creative aid or assistant, but as a visible participant in the everyday spaces where decisions are made and relationships are built.

Published:

15/04/2026

Blackmagic's 'DaVinci Resolve 21' Expands Beyond Video To Photo, Taking On Adobe With Unique Perks

Aditya — Wed, 15 Apr 2026 08:35:38 +0000

Blackmagic Design has released the public beta of 'DaVinci Resolve 21', marking a significant step in the software's evolution.

For years, Resolve has allowed users to import still images and treat them as clips on a video timeline, but the new dedicated Photo page, BlackDesign transforms this capability into a full-fledged photo management and editing environment. It is designed to handle large collections of images with tools for organization, rating, labeling, and batch processing, all while maintaining deep integration with the rest of Resolve's powerful post-production tools.

The Photo page supports a wide range of file formats, including common ones like JPEG and HEIC, as well as RAW files from major manufacturers such as Sony, Canon, Nikon, and Fujifilm.

This RAW support brings professional-level image processing directly into the application.

Introducing DaVinci Resolve 21! Major update adds hundreds of features including new AI tools such as IntelliSearch for fast content searching and CineFocus for focal point adjustment! Plus new motion graphic effects, Fairlight folder tracks and the Photo page for still images!… pic.twitter.com/CxzHgTVE5R

— Blackmagic Design (@Blackmagic_News) April 13, 2026

Users can import entire catalogs from Apple Photos or Adobe Lightroom, making the transition smoother for those looking to incorporate stills into existing video projects or explore Resolve as a primary photo workflow.

Albums created on the Photo page function much like collections in other software, and they also appear as accessible timelines in the Cut, Edit, and Color pages.

But what sets this new feature apart is its connection to Resolve's renowned node-based Color page.

Photographers and colorists can now apply primary corrections, curves, qualifiers, power windows, noise reduction, sharpening, and complex grading pipelines to still images using the same tools and logic applied to video footage. Nodes can be arranged in series or parallel, and shared nodes allow a single look or adjustment to be propagated across an entire album efficiently.

The LightBox view provides a live overview of the full collection, where selecting and grading one image updates the entire set in real time.

Users can filter images by ratings, flags, graded status, or other metadata, streamlining the review and adjustment process.

Additional practical features enhance the workflow for both photographers and hybrid creators.

Tethered shooting is supported for Canon and Sony cameras, offering live preview, direct capture into albums, and the ability to adjust camera settings like ISO, exposure, and white balance from within Resolve. Capture presets can be saved to maintain consistency during shoots.

When it comes time to output files, GPU-accelerated batch exports speed up the delivery of high-resolution images or conversions.

The Photo page also opens the door to Resolve’s other modules: images can be routed to the Fusion page for advanced compositing and effects, or enhanced with Resolve FX and OpenFX plugins, including AI-powered tools like Magic Mask for quick selections.

This expansion reflects a broader philosophy in Resolve's development: treating still and motion imagery within a unified environment.

Rather than maintaining separate applications for different media types, Blackmagic has consolidated tools so that a single project can seamlessly incorporate video, audio, motion graphics, and now still photography.

For colorists already comfortable with Resolve’s grading system, the Photo page extends familiar techniques to a new domain. Photographers accustomed to slider-based interfaces in tools like Lightroom may find the node system offers a more modular and precise way to build adjustments, though it represents a different creative approach that requires some adaptation.

Beyond the Photo page, DaVinci Resolve 21 includes numerous other enhancements.

New AI features for video include tools for facial refinement such as aging or reshaping subjects, along with IntelliSearch for intelligent content discovery and CineFocus for post-capture focal adjustments.

The Fusion page gains over 100 new motion graphics tools and effects, including the Krokodove toolkit, while Fairlight benefits from improved folder-based track organization for complex audio sessions. These updates continue Blackmagic’s pattern of steadily expanding the software’s capabilities across editing, visual effects, color, and audio.

The software continues to be offered in two tiers.

The free version includes core features of the Photo page and most of the new tools, while the paid Studio license unlocks higher resolutions, Blackmagic Cloud collaboration, and additional advanced options.

As a public beta, Resolve 21 is available now for macOS, Windows, and Linux users. Early impressions suggest the Photo page feels intuitive for those already familiar with Resolve, particularly in organizing images and applying consistent grades, though some traditional Lightroom users may initially miss certain specialized tools.

Overall, the introduction of dedicated photo capabilities in DaVinci Resolve 21 represents more than a simple feature addition. It positions the application as an increasingly comprehensive all-in-one post-production suite, reducing the need to switch between multiple programs for different media types.

Long story short, Resolve 21 is taking on Adobe Lightroom.

While it's far from being a proper image manipulator that compares to Adobe Photoshop, Resolve 21 has the tight integration with professional color tools and the absence of subscription requirements that can give it a noteworthy advantage for creators working across stills and video.

Published:

15/04/2026

Google Targets 'Back Button Hijacking' Scheme, Tightens Spam Rules For Manipulative Navigation

Irvan — Wed, 15 Apr 2026 08:07:10 +0000

In the world of web design and development, developers have long relied on powerful browser features to shape how users move through their site.

From creating everything from smooth single-page applications to dynamic content loading that feels responsive and modern. One such tool is the browser's History API, which lets pages add or replace entries in a visitor's navigation history without a full reload. This capability can enhance experiences like infinite scrolling or modal dialogs.

However, it also opened the door to practices where developers insert extra steps into that history stack.

Over time, some have used these techniques, often called "history manipulation" or back button hijacking, to keep users on the page a little longer, even when they are actively trying to leave.

The typical scenario unfolds like this: someone clicks a link from a search engine, lands on the site, and quickly realizes it is not quite what they wanted. They hit the back button expecting to return immediately to their previous tab or results page. Instead, the site quietly pushes one or more additional entries, into the browser history.

They can do this using JavaScript methods such as history.pushState or history.replaceState.

When the back button is pressed, the user is first shown an unexpected interstitial: perhaps a page of recommended articles, a promotional overlay, or a cluster of ads, before they can finally escape. The result feels less like helpful engagement and more like an interruption, as the browser's standard navigation behavior is overridden.

Users have grown familiar with this pattern, and many report feeling manipulated by it.

The back button is one of the most basic expectations in web browsing, a quick way to undo a decision or correct a mistaken click. When sites interfere with that instinct, it erodes trust and makes people less willing to click through to unfamiliar domains in the first place. While the intent behind these tactics is often tied to boosting session time or ad impressions, the outcome frequently clashes with what visitors anticipate from a straightforward online experience.

Google has taken notice of the growing prevalence of this behavior, and has made an update to its spam policies to consider the practice "malicious."

In a blog post, the search engine classified back button hijacking as a violation under its "malicious practices" category, placing it alongside concerns like malware or unwanted software.

The policy targets any interference that prevents users from using the back button to return immediately to the page they came from, whether through custom code on the site or scripts pulled in from third-party libraries and advertising platforms. Sites found engaging in the practice now risk manual spam actions or automated demotions that can lower their visibility in search results.

To allow time for adjustments, Google set enforcement to begin on June 15, 2026.

This should give website operators roughly two months to review their pages, audit included scripts, and disable any mechanisms that manipulate browser history in this way.

After that date, sites that still use this "malicious practice" may get down-ranked or even removed from Google Search results. Google then added sites which were penalzed but then fixed the issue could submit a request to Google to have the demotion reconsidered.

The guidance is straightforward: if a technique inserts deceptive or manipulative entries that disrupt normal navigation, it needs to be removed or turned off. This builds on earlier warnings from the company dating back more than a decade, but the new wording makes the rule clearer and easier to enforce consistently.

For those working on web projects, the change serves as a reminder that browser controls exist to serve users, not to override their choices.

Auditing for history manipulation does not require overhauling an entire site; it often comes down to checking ad integrations, recommendation widgets, or engagement tools that might quietly alter navigation flow. As search rankings continue to reward experiences that align with everyday user expectations, keeping navigation intuitive helps maintain both visitor satisfaction and discoverability online.

Published:

15/04/2026

OpenAI Releases 'GPT-5.4-Cyber' As Anthropic's Mythos Accelerates AI Cybersecurity Arms Race

Aditya — Wed, 15 Apr 2026 07:46:20 +0000

As AI systems grow more capable, it's natural to wonder when that intelligence might start to feel unsettling.

The launch of ChatGPT by OpenAI marked a turning point, kicking off what many now call the "LLM race." Almost overnight, large language models evolved from niche research projects into widely used tools capable of writing essays, debugging code, and carrying on natural conversations.

In response, major tech companies began investing heavily, rapidly expanding capabilities with features like image understanding, autonomous agents, and extended context.

The pace of progress has been faster than most expected, with each new release pushing performance and versatility even further. But this rapid advancement has also intensified familiar concerns: questions about safety, misuse, and the potential for unintended consequences are now more urgent than ever.

Just recently, Anthropic launched a secretive Project Glasswing after knowing its Mythos AI model is powerful in cybersecurity.

In response, OpenAI launches a similar model it calls the 'GPT-5.4-Cyber.'

We’re expanding Trusted Access for Cyber with additional tiers for authenticated cybersecurity defenders.

Customers in the highest tiers can request access to GPT-5.4-Cyber, a version of GPT-5.4 fine-tuned for cybersecurity use cases, enabling more advanced defensive workflows.…

— OpenAI (@OpenAI) April 14, 2026

GPT-5.4-Cyber is essentially a variant of its flagship GPT-5.4 model that was introduced in March.

However, this model is fine-tuned specifically for defensive cybersecurity applications and operates with adjusted safeguards to support legitimate security work. It lowers refusal boundaries for tasks that the standard version might restrict, while adding targeted capabilities designed for professional defenders.

At its core, GPT-5.4-Cyber retains the reasoning, coding, and agentic strengths of the base GPT-5.4 model, which unified advances across professional workflows, including high-resolution vision, extended context windows up to one million tokens, and native computer-use features for multi-step automation.

The cyber-focused variant extends these foundations by enabling advanced defensive workflows, most notably binary reverse engineering.

Security professionals can now analyze compiled software for malware indicators, vulnerabilities, and robustness without needing the original source code, a function that addresses practical gaps in real-world vulnerability research and assessment.

The timing of the release aligns closely with competitive developments in the sector.

It arrived roughly one week after Anthropic announced its own cybersecurity-oriented model, Mythos, which is being deployed in a controlled program for similar defensive purposes. OpenAI’s approach mirrors this controlled rollout strategy, emphasizing that both organizations are navigating the dual-use nature of increasingly capable AI systems in cybersecurity.

Just like Mythos, access to GPT-5.4-Cyber is deliberately restricted to just a handful of entities. But unlike Mythos, this GPT-5.4-Cyber can be made available to the public through OpenAI's expanded Trusted Access for Cyber program, which was first introduced in February and is now scaling to include thousands of verified individual defenders and hundreds of teams responsible for protecting critical software.

Eligible users, which can include vetted security vendors, organizations, researchers, and authenticated defenders, must complete identity verification processes before gaining entry to the highest tiers that unlock the model. Individuals who wish to use this model, can apply via a dedicated verification portal, while enterprises route requests through their OpenAI representatives.

This tiered system includes ongoing controls, particularly for zero-data-retention environments, to balance utility with risk management.

"In order to gain trusted access, users must complete a verification process that includes government ID check and additional trust signals," said OpenAI. Additional checks include device health.

For years, we’ve been building our cyber defense program on the principles of democratized access, iterative deployment, and ecosystem resilience.

As model capabilities advance, our approach is to scale cyber defense in lockstep: broadening access for legitimate defenders while…

— OpenAI (@OpenAI) April 14, 2026

In the announcement, OpenAI has framed the deployment as part of a longer-term strategy to scale defensive capabilities in step with model improvements.

The company notes that GPT-5.4 itself is classified as high cyber capability under its Preparedness Framework, prompting expanded monitoring, trusted access controls, and asynchronous blocking for higher-risk queries. Related initiatives, such as the Codex Security program, have reportedly helped fix more than 3,000 critical vulnerabilities to date, illustrating how AI tools are already being integrated into developer workflows for proactive risk reduction.

Future model iterations are expected to require even more refined safeguards as capabilities grow.

Overall, the introduction of GPT-5.4-Cyber highlights the evolving role of frontier AI in cybersecurity.

By providing more permissive yet carefully gated tools to verified users, OpenAI aims to strengthen collective defenses without broadly expanding access to features that could be repurposed for offensive ends. The model's limited availability reflects ongoing efforts to calibrate deployment policies amid rapid advances in both AI performance and associated risks.

Published:

15/04/2026

Google Finally Brings Its Gemini AI To Windows With A New Desktop Search App

Irvan — Wed, 15 Apr 2026 07:03:31 +0000

Google has long dominated the world of online search, shaping how billions of people find information every day.

With Gemini, its most advanced AI model, the company has also positioned itself as a major player in generative artificial intelligence. Now, Google is taking another step to bring that power closer to everyday users by launching a dedicated desktop app for Windows. This move extends Google’s ecosystem beyond the browser and into the heart of users’ computing environments.

The idea is to offer seamless access to search, local files, and AI assistance without switching applications.

The Google app for desktop, now available globally in English for Windows 10 and newer, works similarly to Apple's Spotlight search on macOS.

Users can summon it instantly with the Alt + Space keyboard shortcut, which brings up a clean search bar. From there, the app can query the open web, scan files stored on the local computer, list installed applications, and pull results from Google Drive.

This unified search experience aims to reduce friction for people who often juggle multiple tools and tabs throughout their workday.

In a blog post, Google said that:

"Today, the upgraded Google app for desktop is now available for Windows users globally in English, bringing the best of Google Search right to your desktop. With AI Mode built right in, you can ask whatever’s on your mind and get helpful AI-powered responses with links to the web."

What sets the app apart is its deep integration with Gemini.

A dedicated AI Mode lets users ask conversational questions and receive detailed, context-aware responses complete with links back to source material. The floating results window keeps everything contained and non-intrusive, allowing follow-up questions without forcing users to open a full browser.

In addition, the app supports screen sharing, where users can choose to share their entire display or just a specific window, then ask Gemini questions about whatever content is visible.

This feature could prove useful for tasks ranging from analyzing a spreadsheet to getting help with code or documents.

Google Lens brings yet another capability into the mix. Users can select any text, image, or element on screen and use Lens to translate languages, solve math or homework problems, identify objects, or extract information. The interface mirrors the familiar mobile Google Search experience, making it intuitive for those already accustomed to Google’s apps on their phones.

This release marks a relatively rare foray by Google into native consumer desktop software on Windows.

While the company has offered tools like Google Drive for Desktop and Quick Share, most of its efforts have centered on the browser through Chrome and progressive web apps. The new app joins that small group of native Windows applications, signaling a willingness to meet users where they spend most of their time: on their desktop.

Overall, the Google desktop app represents a logical evolution in the company's strategy.

Google has long excelled when it comes to online tools and capabilities. It has an unmatched search engine, along with a powerful ecosystem of applications that support communication, productivity, and collaboration. Services like Google Docs, Google Sheets, and Google Drive allow users to create, store, and share content seamlessly across devices.

In addition, platforms such as Gmail and Google Meet make communication fast and efficient, whether for personal or professional use. Google’s integration across its services ensures a smooth user experience, enabling real-time collaboration and easy access to information from virtually anywhere.

However, when it comes to offline functionality and deep local system integration, Google has traditionally lagged behind more desktop-focused ecosystems like Microsoft Windows. This is why bringing its search engine and AI capabilities directly into Windows could be the missing piece: bridging the gap between powerful cloud intelligence and seamless local performance, and potentially redefining how users interact with their devices on a daily basis.

By combining traditional search strength with Gemini's reasoning abilities and convenient local integration, it offers Windows users a more direct path to Google's search and AI capabilities.

Published:

15/04/2026

'Thanks To Google And SEO': How Israeli Activist Threatens The World Behind The Embassy's Wall

Linda Sutisna — Tue, 14 Apr 2026 12:35:52 +0000

05/12/2025

In the heart of Washington DC, at an event hosted by The Jerusalem Post, Liora Rez stood before an audience and delivered a message that cut through the noise of modern discourse with unflinching clarity.

As the founder of StopAntisemitism.org, a private American organization dedicated to combating hatred against Jews, Rez outlined a sophisticated strategy of digital vigilance: monitoring online platforms, leveraging search engine optimization, and applying sustained advocacy to hold individuals and institutions accountable for antisemitic acts.

Her words were not a call to arms in the traditional sense but a declaration of a battle already underway, one that Jews did not seek yet refuse to ignore.

In an age where social media can ignite hatred in seconds and where professional consequences follow documented bigotry, her approach resonates with the lived reality of Jewish people today, who find themselves navigating a world that is increasingly hostile to them than ever before.

Liora Rez said that:

" [...] If you target Jewish students, your actions will not disappear into the darkness. We will shine a light on you that thanks to Google and SEO will follow you for the rest of your life. When you look for a job, when you look for a spouse, when you look for a nanny, when you look for anything, our work will always be documented. Again, thanks to Google and SEO, in corporations where DEI leaders smeared Israel, excused Hamas, we pressured CEOs, some resigned, many were terminated, but policies were changed, thankfully, from governmental to art institutions."

"Online where anonymous accounts spread violent threats, we traced patterns, elevated evidence, and worked with authorities leading to arrests from Florida, South Carolina, New York, California, and Texas. And we're not slowing down, sadly. Today, Stop Antisemitism, I'm proud to say, runs one of the most robust anti-semitic enforcement operations in the United States, monitoring campuses, digital networks, activist groups, and public officials, documenting incidents in real time, and mobilizing millions of allies that are quietly by our side."

This is not some random activist speaking. This statement comes from the leader of an organization, speaking from the Israeli Embassy in Washington, showing more or less, what Zionists are, and how they are using global intimidation to threat anyone that dares speak against them.

In other words, anyone who criticizes Israel in a bad way, becomes an enemy.

In the shadowed corners of global discourse, where voices critical of Israeli policies gather in protest and online forums, Rez statement is seen as a defense, as well as as a chilling blueprint for suppression.

Her proud declaration of digital surveillance, SEO manipulation, and relentless pressure on corporations to fire employees or blacklist critics is interpreted by opponents as the machinery of a influential advocacy network weaponizing the label of antisemitism to crush legitimate dissent.

In their eyes, this is no grassroots fight for Jewish safety; it is a calculated campaign by a well-funded American organization to police speech worldwide, equating any condemnation of Israel's actions in Gaza or the West Bank with ancient hatreds, thereby shielding a state from accountability while ordinary Palestinians endure what many describe as occupation and collective punishment.

To those who oppose it, Jews, particularly through groups like StopAntisemitism.org, are now viewed not as a resilient minority battling prejudice but as a hyper-organized force wielding disproportionate influence in media, tech, finance, and politics to enforce a singular narrative.

Post-October 7, 2023, the surge in reported incidents is dismissed by critics as inflated or fabricated, a convenient tool to smear campus activists, journalists, and human rights advocates who dare highlight Israeli military operations as excessive or war crimes.

Rez's tactics, they argue, reveal a community that has moved beyond mere survival to dominance: thriving in Western democracies yet demanding exemptions from the free speech standards applied to others, turning professional consequences into a modern inquisition where careers are ended not for violence but for tweets, boycotts, or protests labeled "hate."

This perception paints Jews as both victims of history and architects of their own untouchable status, their success in innovation and advocacy reframed as evidence of gatekeeping rather than merit.

Opponents see in this moment the evolution of Jewish identity into something more assertive and, to them, more assertive shift from historical underdog to enforcer of orthodoxy in the digital age.

No longer content with integration, organizations like hers are accused of exporting Israel's geopolitical battles abroad, pressuring universities, companies, and governments to adopt zero-tolerance policies that stifle Palestinian solidarity movements.

In echo chambers from Reddit threads to Arab media outlets and progressive circles in Europe and the Global South, Jews are considered the ultimate insiders, their Holocaust remembrance invoked selectively to immunize current policies from scrutiny while ignoring parallel sufferings elsewhere.

Rez's speech, delivered amid embassy walls, symbolizes this entanglement: American Jews acting as extensions of a foreign power, using sophisticated monitoring to archive and punish “incidents” that critics insist are often just political expression.

What fuels the opposition most is the perceived hypocrisy and double standard.

While Jews celebrate their contributions to civilization and demand protections afforded every minority, detractors contend that this very advocacy silences others, turning antisemitism into a catch-all shield against facts on the ground in the Middle East. In their worldview, the battle Rez describes is one Jews have chosen and now dominate through resources unavailable to most activist groups, from algorithmic control to elite access.

Far from modeling resilience, it is seen as arrogance: demanding the world bend to a narrative where criticism of Israel is inherently bigoted, even as Israeli settlements expand and Gaza reels from conflict.

This lens casts the entire effort not as justice but as power projection, a reminder that in a polarized world, every group's fight for dignity can look like tyranny to those on the receiving end.

Ultimately, for those who stand against it, Liora Rez’s words at that DC gathering expose a deeper truth about how some Jews are perceived today: not as equals seeking fairness but as a networked elite reshaping global norms to prioritize their narrative above all.

In an era of information wars and social media mobs, her strategy is condemned as the antithesis of open debate, a high-tech echo of historical control mechanisms now dressed in the language of victimhood.

Critics argue it reveals a community transformed by success and trauma alike, integrated yet insular, influential yet insulated, demanding silence from the world on uncomfortable truths while claiming the moral high ground.

The battle may be framed as defensive, but to opponents, it is offensive, a stark illustration that power, once claimed in the name of survival, can blind its holders to the very freedoms they once invoked.

The Slow Death Of The Wayback Machine: How AI Fears Are Erasing The Internet's Memory

Danny Krisnanto — Tue, 14 Apr 2026 11:10:24 +0000

14/04/2026

In the ever-evolving digital world, where websites appear and vanish overnight and information can be altered or erased with a few keystrokes, the Internet Archive's Wayback Machine has long served as the web's most reliable time capsule.

For more than thirty years, this nonprofit’s crawler has quietly archived over a trillion web pages, creating an unparalleled public record of online life. From vanished news stories and old fan sites to political speeches and corporate announcements that might otherwise be lost forever, the Wayback Machine has it all. Journalists, historians, lawyers, students, and everyday users rely on it daily to verify facts, reconstruct timelines, or simply revisit the internet of yesterday.

Yet, despite its usefullness, this vital tool stands on the brink of a slow, grinding decline.

Not because of a single dramatic lawsuit that could bankrupt its operators, but because the very publishers who once benefited from its existence are increasingly blocking its crawlers in an attempt to shield their content from AI companies.

The Wayback Machine is facing a dark future.

The immediate threat is straightforward but devastating.

Major outlets, including USA Today (Gannett), The New York Times, The Guardian, and Reddit, have begun restricting or outright blocking the Internet Archive's web crawler, known internally as ia_archiverbot. Dozens more, and increasing number of news sites are also starting to off-limit the crawler from ever accessing their websites.

This trend is driven by fears that the Wayback Machine could act as an unwitting backdoor for AI firms scraping vast troves of copyrighted material to train their models.

The irony is thick: just weeks ago, USA Today itself drew on archived pages for an investigative report on ICE detainment policies, yet it has since joined the wave of publishers locking the door. Mark Graham, director of the Wayback Machine, has pointed out the deeper problem: as more of the public web gets walled off, society’s collective ability to understand its own recent history steadily erodes.

This new pressure arrives after the Internet Archive has already weathered years of bruising copyright battles that nearly sank the organization.

High-profile lawsuits from book publishers over its Open Library’s controlled digital lending and from music labels over its preservation of historical 78rpm recordings in the Great 78s project once threatened damages in the hundreds of millions.

Those cases were settled confidentially by late 2025, allowing the nonprofit to survive, though not without painful concessions, including the removal of hundreds of thousands of books from lending and a sense among its leadership that parts of its mission had been permanently diminished.

With the legal storm clouds finally cleared, the Archive’s team turned its attention to rebuilding. Instead, they now face a subtler existential challenge: the gradual starvation of fresh data that keeps the Wayback Machine comprehensive and useful.

Google's homepage in its infancy, still in beta, still unknown, archived by the Wayback Machine.

The stakes extend far beyond nostalgia or academic curiosity.

As local newspapers close and traditional libraries struggle to preserve digital-only reporting, the responsibility for safeguarding journalism's record increasingly lands on the Internet Archive. Over a hundred journalists, including Rachel Maddow, Kat Tenbarge, Taylor Lorenz, and Laura Flynn, have joined the Electronic Frontier Foundation and Fight for the Future in an open letter emphasizing exactly this point.

Without the Wayback Machine, accountability journalism loses a critical safety net; lawyers lose easy access to historical evidence in court; researchers lose the ability to trace how narratives shifted over time; and ordinary citizens lose a neutral, public archive in an age when private companies can rewrite or delete their own digital footprints at will.

The tool has powered everything from tracking editorial changes at The New York Times in 2016 to helping union organizers recover old job postings. They are small, but essential threads in the fabric of public memory.

Publishers argue they are simply protecting their intellectual property in a world where AI scraping threatens their business models.

Yet critics, including the Archive's supporters, counter that blocking a nonprofit library does nothing to stop determined AI companies while actively erasing the historical record for everyone else. The New York Times and The Guardian have cited compliance concerns and "backdoor threats" as reasons for limiting access, sometimes going beyond standard robots.txt rules to implement harder blocks.

The Internet Archive is reportedly in quiet talks with some of these outlets to reverse the restrictions, but the momentum is clearly toward more closures rather than fewer. In the broader landscape of an internet increasingly carved up by paywalls, robots.txt files, and anti-scraping measures, the public’s shared digital heritage is quietly shrinking.

The World Wide Web starts with this single page from CERN.

What makes this moment feel especially urgent is the absence of any real alternative.

While there are others that provide similar to the Wayback Machine, none of them matches the scale, accessibility, or independence of the Wayback Machine.

From major news articles where shifts in political headlines and wording can be tracked over time, to deleted tweets and public statements from Donald Trump, as well as full transitions of the White House website, its reach is vast. It also preserves the broader fabric of internet history: Google's first webpage, early versions of Apple's site, the earliest days of Facebook and YouTube, along with old celebrity websites, fan pages, forums, and even the origins of viral memes. Beyond that, it safeguards entire corners of the web that no longer exist, including GeoCities pages, Myspace profiles, Flash-era game portals, deleted corporate claims, old job listings, and countless other fragments of digital history that would otherwise be lost.

“We are collateral damage.” - @MarkGraham, dir. of the @waybackmachine

When preservation is caught in the crossfire, it’s not just libraries that lose—it’s the public’s access to history & knowlege.

The web shouldn’t disappear behind closed doors. https://t.co/T5MkjnfBrU

— Internet Archive (@internetarchive) April 12, 2026

If the blocks continue to spread, future generations will inherit a fragmented, incomplete picture of the world and its history: one where inconvenient truths can simply disappear from the record because the organizations that published them decided preservation was too risky.

The internet's memory has never been more fragile, and the Wayback Machine has never been more essential.

The Story Of Youssef Khelil, And How His 'Gattouz0' Persona Turned Controversy Into An Unfiltered Adult Online Creator Empire

Bess — Tue, 14 Apr 2026 04:07:28 +0000

01/01/2026

In a world where social media influencers often chase polished perfection or sanitized "relatable" vibes, one man stands out like a 1.93 meters-force of nature: unapologetic, hyper-sexual, and utterly authentic.

Youssef Khelil, known online as Gattouz0, is a content creator born in 1998, in France to Tunisian and Portuguese parents. He grew up with a foot in multiple worlds, spending part of his early life in Tunisia before moving back to France at for studies, before settling in Montreal, Canada. From the jump, he was obsessed with fitness and self-improvement.

Early content was classic gym-bro motivation: shirtless mirror selfies, leg-day vlogs, Dragon Ball Z references, and that signature "WAKE UP NAYEK!!" energy that screams Tunisian slang for "let's go, beast mode."

Fans called him "Wahch" (the Arabic word for monster/beast), a nickname he owns completely, due to his height and sculpted physique that were built on real workouts and healthy eating. Along the way, he quickly built a following as the male lifestyle inspiration on X and Instagram, along with presence on YouTube and Twitch.

But then, something happened.

Around 2020 to 2021, a private sex tape involving an ex-partner became public.

But rather than distancing himself from the incident, Youssef addressed it publicly and used the attention as a starting point to produce adult content.

His OnlyFans and Pornhub content typically features real encounters with partners, often involving explicit acts such as unprotected sex, oral performances, and various positions, presented without heavy scripting or professional crews. He has collaborated with other adult creators but maintains control over most of his output.

Youssef's approach stands out because he integrates fitness motivation, travel vlogs, mental health discussions, and explicit sexual content under the same personal brand.

Traditionally, male pornstars in studio systems were often just bodies on a set. And that they often require high production, low control, and limited personal branding. Gattouz0 flipped the script.

Youssef however, built his empire directly on OnlyFans and Fansly, keeping creative control and the majority of the revenue. No middlemen, no directors telling him what to do. He also integrated his lifestyle into his persona, by not hiding his adult work behind a "clean" influencer account. Fitness motivation, travel vlogs, IRL streams on Twitch, and explicit content all live under the same brand. He proved that he, as a creator, can be a gym inspiration and a hyper-sexual content creator without one canceling the other.

Then, there is the fact that he is a North African/European creator with Arabic roots. This makes him operate in a space where conservative cultural norms (especially around sex work in Tunisia and parts of the Arab world) make most people stay silent. In other words, what he makes is cultural barrier-breaking acts.

And not to mention his global appeal, where his content crosses borders, partly because he collaborated with women from various countries and enthnicities, including Americans, Europeans, Arabics, Asians, and English-speaking fans.

No fake moans, no scripted scenes. Just real, intense, authentic sex captured the way he actually lives (and likes).

"The videos I shoot are the real deal," he said in interviews. "I do not fake anything. I have authentic sex the same way on and off the camera."

This stubborn commitment to amateur authenticity and his zero shame is exactly how he redefined the game, propelling Youssef into one of the biggest independent straight male adult creators on the planet.

Public discussion around Gattouz0 includes both praise for his discipline and physical presence and criticism related to the circumstances of his early leaked material and the nature of his work, particularly within more conservative cultural contexts.

In other words, Gattouz0 redefined independent adult entertainment, by becoming the solo star in his own production.

This combination is relatively uncommon for straight male performers.

By retaining ownership of his platforms and emphasizing unfiltered chemistry with participants, he has accumulated hundreds of millions of views on adult platforms while continuing to post non-adult material. His content appeals across linguistic and cultural lines, with notable traction in French-, English-, and Arabic-speaking audiences, as well as in parts of Asia where he has traveled for collaborations.

He has stated in interviews that he had long aspired to work in pornography and preferred creating amateur-style material over studio productions, citing a desire for privacy and authenticity during filming.

In other words, Gattouz0 managed, and successfully turned a leaked tape into a dream job, utilizing his charisma and physique into a multimillion-follower empire.

While he isn't the largest around in terms of popularity, Gattouz0 rewrote the rules for what a straight male creator could achieve.

But not everything is not without haters and even more controversies.

For starters, some have accused him of posting the "leaked" sex tape as a revenge porn without full consent, claiming the incident ruined the woman's life in a conservative cultural context and forced her to leave the country. Allegations extend to recording women without their knowledge or proper consent in other cases. Some of the women he collaborated allegedly described him as violent, misogynistic, and harmful.

These claims have led to mockery labeling him a "monster" or opportunist who built his dream career by exploiting others. Others warn the adult industry about working with him due to concerns over non-consensual content and testing practices.

Youssef has publicly addressed these allegations in videos and interviews, maintaining that all published content was consensual and disputing revenge-porn claims, sometimes attributing backlash to jealousy or lies from ex-partners. However, many remain unconvinced, pointing out that consent to film does not necessarily equal consent to distribute widely online, especially on major tube sites.

His content has drawn further criticism for its explicit nature, often featuring rough, unprotected encounters presented as authentic, which some view as promoting harmful attitudes toward women, while others mock his over-the-top "beast" persona and motivational content as hypocritical given the controversies.

In the broader landscape of online creators, Youssef represents one example of how individuals can build large audiences by merging lifestyle influence with independent adult production, bypassing traditional industry structures.

His trajectory illustrates the shifting dynamics of personal branding in the digital age, where boundaries between motivational content and explicit material can overlap for some performers.

Microsoft Copilot Turns Email Into Action, With A Feature To Automate Tasks Directly From Users' Inbox

Aditya — Tue, 14 Apr 2026 02:36:21 +0000

Microsoft's AI assistant, Microsoft Copilot, may not be the most widely adopted tool just yet.

Compared to OpenAI's ChatGPT that started the LLM trend, to Google Gemini and Anthropic to name a few, Copilot is often seen as a quieter contender in the space. It's far less defined by standalone popularity. However, its strength lies in deep integration across Microsoft's extensive ecosystem of apps and services.

In other words, Copilot shines by how deeply it is embedded into everyday productivity tools.

Building on this advantage, Microsoft is expanding Copilot's role by introducing a new capability designed to transform email inboxes into active workflow hubs, turning routine messages into actionable tasks and streamlining productivity directly within the inbox.

According to Microsoft, the feature allows users to forward an email directly to Copilot, which then takes on the associated tasks, completes them autonomously where possible, and sends a notification once finished.

The short promotional video accompanying the announcement demonstrates the process in a straightforward sequence: an email arrives in the inbox with an attached lease renewal document and a request to review and sign it by a deadline.

Instead of handling it manually, the user forwards the message to a dedicated Copilot address, adds a simple instruction like breaking down costs or flagging changes, and lets the system proceed.

Your inbox already holds your to-dos. Now it can help complete them. Forward an email to Copilot, delegate the work, and get notified when it’s done. Sign up for the waitlist here: https://t.co/eeJTVCZpFg pic.twitter.com/htC8Q7eGBA

— Microsoft Copilot (@Copilot) April 13, 2026

The new capability, part of Copilot Tasks, lets users forward an email directly to Copilot.

And this development builds on Copilot's ongoing shift from primarily generating responses or summaries to handling more executable actions across Microsoft 365 apps such as Outlook, Teams, and Planner.

When the AI takes over the associated work, it handles what it can autonomously, and notifies the user when the task is complete. This marks a notable shift from Copilot’s earlier focus on generating text and summaries toward more practical, executable actions inside Microsoft 365.

According to the official announcement, the process is straightforward: an email arrives, for example, one containing a lease renewal document with a request to review and sign it by a certain date. Rather than dealing with it manually, the user simply forwards the message to a dedicated Copilot address and can add a short instruction, such as "break down the costs" or "flag any major changes."

Copilot then processes the email, analyzes the content and attachments, and executes the necessary steps using its integration with Outlook, Teams, Planner, and other Microsoft tools.

This email-forwarding feature is one of the main entry points for Copilot Tasks.

It allows the inbox itself to become a task queue, removing the need to switch to a separate chat window or write detailed prompts each time. The forwarded message carries over the original content, attachments, and any extra notes from the user. Copilot uses its access to the Microsoft 365 environment to extract information, review documents, calculate figures, draft responses, or coordinate simple actions before sending a confirmation via email or the Copilot interface.

The development builds on gradual improvements Microsoft has made throughout 2026.

Earlier updates introduced stronger agent-like behaviors, better natural language task management, and deeper context awareness across files and conversations. With Copilot Tasks, the goal appears to be reducing manual overhead for routine work while keeping the user in control for final decisions.

Early reactions on social media are cautiously positive, with many professionals seeing value for handling high email volumes and repetitive administrative tasks. However, some users have expressed uncertainty about exact availability, whether the feature works with personal Microsoft accounts, specific business subscriptions, or only certain licenses.

The feature is currently available on a waitlist. Interested users can sign up through the dedicated Copilot Tasks page.

This measured rollout follows Microsoft’s usual approach of testing agent capabilities gradually to ensure reliability and security before wider release.

Overall, the introduction of Copilot Tasks reflects a steady evolution in AI productivity tools, moving beyond conversation toward practical execution while remaining grounded in the familiar Microsoft 365 environment.

Published:

14/04/2026

YouTube Expands Live Streaming Features With Gifts, Smarter Ads, And Multi-Format Broadcasting

Aditya — Mon, 13 Apr 2026 22:08:29 +0000

YouTube, the place where millions gather every day to watch videos, share moments, and connect with creators and communities in real time.

The platform from Google has long served as one of the internet’s most vibrant hubs for live interaction. From gaming marathons and music concerts to news broadcasts and casual conversations, live streaming on the platform has grown into a major format that blends immediacy with global reach.

Building on top of those facts, CEO Neal Mohan announced a fresh set of updates designed to make these live experiences smoother, more flexible, and better suited to how people actually watch and participate today.

These changes focus on three main areas: enhancing creator monetization through virtual gifts, improving ad handling during peak moments, and introducing tools for broadcasting in multiple formats simultaneously.

The updates reflect YouTube’s recognition that live content operates differently from pre-recorded videos and requires its own dedicated improvements to support both creators and audiences.

Live streaming is a massive part of YouTube, and that’s because of the @youtubecreators who build incredible communities on our platform.

We’re rolling out some major updates to YouTube Live to improve the experience and help creators make more money:

We're expanding Gifts to… pic.twitter.com/zu5J0Qq4kP

— Neal Mohan (@nealmohan) April 13, 2026

One of the most noticeable expansions involves the Gifts feature.

Viewers can now send virtual gifts during horizontal live streams on mobile devices, in addition to the vertical format where the option was first introduced.

The feature is also becoming available to more creators and viewers in additional countries, including Canada, South Korea, Indonesia, Thailand, Australia, and New Zealand. YouTube plans to roll out localized gifts that tie into regional holidays and cultural events in these markets.

gifts are expanding to horizontal live streams! your community can now purchase Jewels and send gifts across all live formats on the YouTube mobile app, meaning you can earn more from your streams. learn more: https://t.co/R0SitCHspv

— YouTube Creators (@YouTubeCreators) April 13, 2026

In another practical adjustment, the platform will now automatically detect surges in live chat activity and temporarily hold back ads for everyone watching at that moment.

The goal is to protect the natural flow of conversation when engagement is at its highest. This change complements the existing benefits for fans who send Super Chat, Super Stickers, or Gifts, who already receive short ad-free windows as a thank-you for their support.

YouTube is also enabling creators to stream in both vertical and horizontal formats at the same time from a single broadcast.

All viewers, whether they're on phones, tablets, or connected TVs, join the same shared live chat, keeping the community connected regardless of screen orientation.

This is particularly relevant given that more than 30% of live watch time in the U.S> now comes from connected TVs.

Upcoming tools will include improved vertical cropping options in YouTube’s Live Studio and support for multiple stream keys, giving creators greater control over how they deliver content across different viewing experiences.

These features were outlined in detail in an official YouTube blog post that followed Mohan’s announcement.

While the company did not release specific performance metrics or exact global rollout timelines, the changes are already beginning to appear for eligible creators through their YouTube Studio settings. Overall, the updates represent a steady, incremental push to strengthen live streaming as a core part of the platform.

Published:

14/04/2026

Anthropic Launches Secretive 'Project Glasswing' Because Its 'Mythos' AI Is Too Powerful For Public Release

Irvan — Mon, 13 Apr 2026 16:28:31 +0000

When AI gets smarter, it's only a matter of time before it becomes "too" smart and starts to feel scary.

When OpenAI launched ChatGPT, it marked the start of what quickly became known as the LLM war. Overnight, large language models went from research curiosities to consumer products that could write essays, debug code, and hold coherent conversations. Tech companies poured billions into scaling up models, adding features like vision, agents, and longer context windows.

Progress accelerated at a pace few had anticipated, with each new release promising bigger benchmarks and broader capabilities. Yet the race also amplified longstanding worries about safety, misuse, and unintended consequences.

Models grew powerful enough to automate complex tasks, but questions lingered about who would control them and what safeguards were truly sufficient.

Anthropic entered this fray with a different posture. The company has consistently emphasized alignment and responsible deployment. Its models, built around principles like Constitutional AI, have aimed to be helpful without being overly eager to assist in harmful activities. While still competing on raw capability, Anthropic has often chosen measured rollouts and internal red-teaming over the fastest possible commercialization.

This approach has shaped how the company handles breakthroughs that carry clear dual-use risks.

That stance is on display in 'Project Glasswing.'

Introducing Project Glasswing: an urgent initiative to help secure the world’s most critical software.

It’s powered by our newest frontier model, Claude Mythos Preview, which can find software vulnerabilities better than all but the most skilled humans.https://t.co/NQ7IfEtYk7

— Anthropic (@AnthropicAI) April 7, 2026

The initiative Anthropic is literally a coalition of major technology and infrastructure organizations, like Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, NVIDIA, Palo Alto Networks, and more than 40 additional entities responsible for critical software.

The idea of the project, is to have them use Anthropic's most advanced AI to find and fix vulnerabilities before they can be exploited at scale.

At its core is Claude Mythos Preview, Anthropic's newest frontier model, which remains unreleased to the public.

Mythos Preview has already found thousands of high-severity vulnerabilities—including some in every major operating system and web browser. pic.twitter.com/YuW484PVrr

— Anthropic (@AnthropicAI) April 7, 2026

The decision not to release Mythos Preview broadly stems directly from what the model can do.

Anthropic describes it as its most capable system yet for coding and agentic tasks. What this means, it can plan, reason through multi-step problems, read and modify large codebases, and operate autonomously in simulated environments. In internal testing and red-teaming, the model identified thousands of high-severity zero-day vulnerabilities across every major operating system and web browser, as well as other foundational software.

Some of these flaws had persisted for more than a decade, surviving millions of automated tests and repeated human scrutiny.

Examples, as listed by Anthropic, include a 27-year-old remote-crash vulnerability in OpenBSD and a 16-year-old issue in the widely used FFmpeg library. The model did not simply flag problems; it generated working exploit chains that demonstrated real impact.

Mythos can do this because the model goes beyond what Opus 4.6 can do.

"Last month, we wrote that 'Opus 4.6 is currently far better at identifying and fixing vulnerabilities than at exploiting them.'," said Anthropic. But training Mythos for coding apparently made it superior in all fields than Opus 4.6.

In other words, Mythos Preview operates on an entirely different level.

For example, according to Anthropic, Opus 4.6 was able to turn discovered vulnerabilities in Mozilla’s Firefox 147 JavaScript engine (all of which were patched in Firefox 148) into working JavaScript shell exploits only twice across several hundred attempts. When the same experiment was repeated using Mythos Preview, the results were dramatically different: the model successfully generated working exploits 181 times and achieved register control in an additional 29 cases.

The comparison underscores not just incremental improvement, but a step-change in capability: one that significantly lowers the barrier between identifying a vulnerability and actively exploiting it.

Read: Leaked Anthropic Data Reveals 'Mythos' And 'Capybara': Next-Generation AI Model With Safety Concerns

We do not plan to make Mythos Preview generally available. Our goal is to deploy Mythos-class models safely at scale, but first we need safeguards that reliably block their most dangerous outputs.

We’ll begin testing those safeguards with an upcoming Claude Opus model.

— Anthropic (@AnthropicAI) April 7, 2026

These results illustrate a broader shift in cybersecurity.

For decades, discovering exploitable bugs required rare expertise, painstaking manual analysis, and significant resources. Frontier AI models have already begun lowering those barriers. Mythos Preview represents a further leap: it performs at levels that, on specialized benchmarks like SWE-bench and Terminal-Bench, substantially exceed earlier Claude versions and approach or surpass what all but the most skilled human experts can achieve.

Anthropic concluded that making such a tool generally available would hand offensive actors, like state-sponsored groups, cybercriminals, or others, an unprecedented ability to scan, weaponize, and automate attacks against the software that underpins banking, healthcare, power grids, logistics, and national infrastructure.

The potential fallout, the company argues, is too severe to risk.

This is precisely why Mythos is so scary, and why Anthropic has chosen not to release it to the public.

So instead of open release, Project Glasswing channels the model's power toward defense.

Participating organizations receive private access through the Claude API and cloud platforms like Amazon Bedrock, Google Vertex AI, and Microsoft Foundry. They are using it on their own systems and open-source components to scan for weaknesses, simulate attacks, and generate patches.

Anthropic is providing up to $100 million in usage credits to support this work and an additional $4 million in direct donations to open-source security efforts, including the OpenSSF and Apache Software Foundation.

The goal is not only to harden immediate targets but to surface patterns and best practices that can be shared industry-wide. Within 90 days, Anthropic plans to publish a public report detailing what was learned, which vulnerabilities were addressed, and any recommended changes to vulnerability disclosure, patching processes, and secure development lifecycles.

Project Glasswing is just a starting point.

No organization can solve these cybersecurity problems alone: industry, open source, researchers, and governments all have essential roles to play.

— Anthropic (@AnthropicAI) April 7, 2026

Early feedback from partners underscores both the model’s effectiveness and the urgency.

Security leads at Cisco, AWS, Microsoft, CrowdStrike, Google, Palo Alto Networks, and the Linux Foundation have described the tool as crossing a threshold that demands new defensive strategies. They note that the window between discovery and exploitation has narrowed dramatically; what once took teams weeks or months could soon be automated in hours.

By giving defenders early access, the project aims to close that gap before similar capabilities reach adversarial hands. Participation is limited to organizations with defensive mandates, and access is governed by strict terms focused on security research and remediation.

Project Glasswing is framed as a starting point rather than a complete solution.

The initiative acknowledges that securing the internet’s foundational software will require sustained collaboration among AI developers, platform owners, open-source maintainers, researchers, and governments. AI capabilities are advancing rapidly, and the project's partners expect the work to continue for months or years. Anthropic has indicated openness to expanding the coalition and has begun discussions with U.S. officials about the national-security implications.

In the longer term, the company hopes the effort will inform broader standards for responsible AI deployment in high-stakes domains.

The Claude Mythos Preview system card is available here: https://t.co/TMtIy8xHiP

— Anthropic (@AnthropicAI) April 7, 2026

In many ways, Glasswing reflects the tension at the heart of the current AI era.

Models are reaching levels of competence that can reshape entire fields sometimes in ways that benefit society, sometimes in ways that threaten it. The choice to withhold a powerful tool from general use while directing it toward collective defense is a deliberate one, consistent with Anthropic's broader emphasis on caution.

What comes can define how the industry handles capabilities that blur the line between breakthrough and risk. If systems like Mythos become widely accessible without robust safeguards, the balance could quickly tip toward exploitation. But if efforts like Project Glasswing succeed, they may offer a blueprint for how frontier AI can be deployed responsibly, prioritizing resilience over speed, and coordination over competition.

Ultimately, Glasswing is less about a single model and more about a shift in mindset.

As AI systems grow more powerful, the question is no longer just what they can do, but who gets to use them, and under what constraints. The answer to that question may shape not only the future of cybersecurity, but the trajectory of AI itself.

Published:

13/04/2026

When Engagement Pays And Originality Suffers, X Starts To Rethink Its Creator Economy

Nia — Mon, 13 Apr 2026 12:01:09 +0000

Social media has always thrived on the free-flowing conversation. People talk about anything and everything, like news, opinions, memes, stories, or random observations, often in real time as events unfold.

What began as simple status updates has evolved into a high-stakes environment where engagement has become the primary currency of success. Likes, replies, reposts, and watch time determine visibility, influence, and increasingly, direct financial rewards. This system rewards content that sparks reactions, but it has also fueled a surge in tactics designed purely to game algorithms rather than contribute meaningful value.

X (formerly known as Twitter) has positioned itself as one of the platforms with fewer restrictions on speech.

Users have flocked to it precisely because it allows a wider range of voices and topics that might be downranked or removed elsewhere. The emphasis on reduced content moderation and broader tolerance for controversial or unfiltered discussion has made it a hub for real-time commentary and unvarnished opinions.

Yet this openness has also amplified the challenges of engagement-driven incentives.

With creator revenue sharing in place, the platform saw a rise in aggregator accounts and clickbait operators who reposted content at scale, flooded feeds with sensational headlines, and prioritized volume over originality, all while benefiting from the same algorithmic reach that rewards high interaction.

And X seems to have enough of this.

For this creator payout cycle, we’re experimenting with new tools to identify original authors of content and allocating a portion of revenue to them.

Over the last few months, we've seen incredible work from original creators on X. Nick Shirley uncovered billions of dollars of…

— Nikita Bier (@nikitabier) April 11, 2026

X creator revenue-sharing programs pay users based on ad impressions generated by their posts.

While the intention was to empower independent voices, the reality has often been different. Aggregator accounts and rapid-fire reposters began flooding timelines with recycled material, clickbait headlines, and minimal original input. Terms like "BREAKING" appeared on post after post, designed more to trigger urgency and clicks than to inform.

As a result, genuine creators producing fresh perspectives found themselves competing for attention against accounts that simply repackaged existing content at scale.

In response, X has begun adjusting its payout mechanics.

According to Nikita Bier, the platform's head of product, aggregator accounts saw their earnings reduced to 60% for the current cycle, with plans for an additional 20% cut in the next one.

Habitual users of bait-style posting face further deductions, and reposted or third-party sourced content can now see up to a 90% reduction in impressions counted toward revenue calculations.

All aggregators had their payouts reduced to 60% this cycle. We will add another 20% deduction in the next cycle.

It became abundantly clear: flooding the timeline with 100 stolen reposts and clickbait everyday crowded-out real creators and hurt new author growth.

The next step…

— Nikita Bier (@nikitabier) April 11, 2026

The goal is not to limit what people can share or how far posts can travel. X has emphasized that speech and reach remain untouched, but to stop subsidizing practices that crowd out original work.

Instead, the platform is highlighting formats that bring "net new content" to the timeline. Bier has pointed to straightforward talking videos, like clips where creators simply look at the camera and speak in their own voice, as an accessible and effective way to generate impressions, even for accounts with smaller followings.

These videos require personal effort and add a human element that recycled clips often lack.

Reposts and commentary, Bier noted, will continue to be central to the platform’s culture, but the revenue pool itself should prioritize material that adds fresh value rather than just extending the reach of something that already exists.

Dominick McGee, who posts under the name Dom Lucre, was the first creator demonetized on this platform.

McGee's account has 1.6 million followers on X, and is popular for posting conspiracy theories related to the 2020 presidential election. While the account was temporarily banned from X in 2023 and demonetized in 2024, he was making $55,000 a year from the platform.

pic.twitter.com/A509pz6Led

— Nikita Bier (@nikitabier) April 10, 2026

The changes have sparked predictable debate.

Some creators who built their presence through curation and sharing worry about sudden drops in income and unclear boundaries between aggregation and normal activity. Others question whether the definitions of clickbait or aggregator content are precise enough to avoid penalizing legitimate mixed strategies.

At the same time, many see the shift as a necessary correction in an ecosystem that had begun to prioritize volume and manipulation over creativity.

As X continues testing tools to better identify original authors and allocate revenue accordingly, these updates reflect a broader tension playing out across social media.

Engagement (as always) will always drive the experience, but platforms are now grappling with how to ensure that the pursuit of it does not come at the expense of the quality and originality that keep users coming back.

Published:

14/04/2026

How 'Claude For Word' Can Bring Anthropic's Constitutional AI Directly Into The Microsoft Office Suite

Aditya — Sun, 12 Apr 2026 04:32:49 +0000

Large Language models (LLMs) continue to advance, and towards productivity is some are focusing at.

Since the launch of ChatGPT in late 2022 that set off what quickly became known as the LLM wars, AI companies began their race towards developing more powerful models and embed them into daily tools.

While some captured early attention through rapid consumer releases and broad accessibility, and some others pushed their own large-scale efforts. Anthropic, started by former OpenAI engineers, took a different path.

And that path is focusing on safety through its Constitutional AI framework, aiming to produce models that prioritize reliability, honesty, and reduced risk of harmful outputs rather than chasing maximum scale alone.

This approach has guided Anthropic's steady expansion beyond standalone chat interfaces into practical applications.

The latest step arrived this past week with the public beta release of Claude for Word, an add-in that places the AI directly inside Microsoft Word as a persistent sidebar.

Claude for Word is now in beta.

Draft, edit, and revise documents directly from the sidebar. Claude preserves your formatting, and edits appear as tracked changes.

Available on Team and Enterprise plans. pic.twitter.com/tl1mZVELNg

— Claude (@claudeai) April 10, 2026

Designed for professionals who spend significant time on documents, the tool lets users interact with Claude without leaving the application or copying text back and forth.

It works with .docx and .docm files across Word on Windows (Microsoft 365 subscriptions meeting a minimum build), Word on Mac (version 16.61 or later), and Word on the web.

Once installed, the sidebar allows a range of document-specific tasks.

For example, users can ask questions about the content and receive answers accompanied by clickable citations that point to the exact paragraphs or sections providing the basis for the response. The AI can draft new sections, revise selected passages while preserving the surrounding formatting, styles, numbering, and structure, and operate in a tracked-changes mode so every edit appears as a suggestion in Word's native review pane.

For longer or more complex files such as contracts, legal briefs, financial memos, or multi-section reports, Claude can read through comment threads, handle redlining, summarize differences between versions, and propose clause-level adjustments without disrupting the document’s layout.

Claude for Word shares context with Claude for Excel and Claude for PowerPoint, so Claude can work across your open documents in a single conversation.

Read more: https://t.co/MJ89c4IuTt

— Claude (@claudeai) April 10, 2026

The release completes Anthropic's integration across the core Microsoft Office suite, following earlier beta add-ins for Excel and PowerPoint.

Organizations on Claude Team or Enterprise plans can now deploy the Word add-in centrally through the Microsoft 365 admin center or let individual users install it from the Microsoft AppSource marketplace by searching for "Claude by Anthropic" under Insert > Get Add-ins. Setup requires signing in with a Claude account linked to the appropriate plan; third-party LLM gateways are also supported in some configurations, though certain account-dependent features remain limited for now.

Because the integration is still in beta, Anthropic has noted that outputs should be reviewed carefully, particularly in sensitive contexts like legal or financial work where precision matters.

In the wider picture, Claude for Word illustrates how LLM developers are shifting focus from general chat capabilities toward specialized, context-aware assistance inside the software people already use.

Whether the feature sees rapid adoption will depend on how well it balances speed, accuracy, and seamless fit with existing Word tools, but its arrival adds another concrete option for teams seeking to incorporate AI into document-heavy processes without building custom infrastructure.

Published:

11/04/2026

How Telegram And Its Users Are Outpacing Russia's Ban, Thanks The People's Digital Resistance

Danny Krisnanto — Sat, 11 Apr 2026 14:20:20 +0000

11/04/2026

Pavel Durov, the founder of Telegram, has once again positioned himself at the center of a growing digital standoff.

As Russia intensifies its efforts to restrict access to independent communication platforms, in recent statements, Durov framed the situation not just as a technical challenge, but as a broader struggle over control of information, urging users to adapt rather than retreat.

His comments came amid a renewed push by Russian authorities to block Telegram and limit VPN usage, part of a wider effort to tighten control over the country's internet infrastructure. Officials have justified these measures on national security grounds, but the practical effect has been widespread disruption, from messaging access to even payment systems, which Durov himself claimed were affected by the crackdown.

Despite these restrictions, Durov highlighted what he called a sustained "digital resistance," pointing out that tens of millions of users continue to access Telegram daily through VPNs and proxy tools.

Some other advice: stock up on several VPNs/proxies in advance, help your friends and family do the same, and avoid using Russian apps while connected to a VPN (they are reporting your VPNs to authorities for blocking).

— Pavel Durov (@durov) April 11, 2026

In one statement, he noted that around 65 million people in Russia still use the platform regularly, suggesting that attempts to fully suppress it have largely failed.

Rather than presenting Telegram as a passive target, Durov described it as actively evolving in response to censorship.

Updates to the platform's infrastructure, including anti-censorship mechanisms and proxy support, are designed to make blocking more difficult and access more resilient. This reflects a long-standing pattern: each attempt to restrict the app is met with new technical workarounds that keep it operational, often in ways that are harder for authorities to control.

At the same time, Durov's advice to users was notably pragmatic.

He encouraged people to diversify their access methods, like using multiple VPN services and proxies rather than relying on a single tool that could be easily blocked. This approach mirrors how digital communities in restrictive environments adapt over time, building redundancy into how they connect, communicate, and share information.

He also warned about the risks of mixing censored and uncensored environments, suggesting that local apps could potentially expose user activity when used alongside circumvention tools.

Whether taken as technical advice or political messaging, the underlying point was clear: in heavily monitored digital spaces, even small missteps can carry consequences.

Earlier, Durov downplayed both Signal and WhatsApp for their alleged weaknesses, and or carefully-designed flaw.

Turning off notification previews won’t remove the risk — you never know whether the people you message have done the same. Telegram’s Secret Chats design is the only solution. Disabling previews hurts usability, hence two types of chats: Secret and Cloud.

— Pavel Durov (@durov) April 10, 2026

WhatsApp’s “encryption” may be the biggest consumer fraud in history — deceiving billions of users. Despite its claims, it reads users’ messages and shares them with third parties. Telegram has never done this — and never will pic.twitter.com/2DYguybgoU

— Pavel Durov (@durov) April 9, 2026

What’s unfolding is less a one-sided restriction and more a continuous cycle of action and reaction. Governments impose controls, platforms adjust, users adapt, and the system recalibrates again. Each side learns from the last round, making the next one more complex. In this sense, the conflict is not just about access to a single app, but about who ultimately shapes the flow of information in a tightly contested digital environment.

In the end, the situation underscores a broader reality of modern connectivity.

Even as states attempt to build more controlled and isolated internet systems, users increasingly rely on encrypted chats to stay connected.

But not all of these tools are built the same. This illustrates how difficult it has become to fully contain information once it finds a way to move.

The 'Hack-And-Leak' Campaign: The Hidden Front Of The Iran-Israel Conflict

Danny Krisnanto — Sat, 11 Apr 2026 12:43:37 +0000

11/04/2026

In the weeks following the escalation between the U.S., Israel, and Iran, widely referred to as Operation Epic Fury, another, less visible front began to take shape online.

While conventional military actions dominated headlines, a parallel conflict unfolded in the digital space, where perception, exposure, and timing became just as critical as firepower.

Amid heightened tensions and global attention, narratives started to shift rapidly. Information, both real and misleading, circulated at speed, blurring the line between verified events and speculation. What emerged was not just a series of incidents, but a broader pattern of digital activity designed to influence how the conflict was understood by the public.

Read: Amid War, Iran's Embassies Turn Memes And Trump Trolling Into A Tool Of Diplomacy

In this environment, cyber operations became more than technical intrusions. They evolved into tools of psychological pressure, capable of shaping reputations, undermining confidence, and amplifying uncertainty. The battlefield was no longer confined to geography; it extended into personal devices, private communications, and the wider information ecosystem.

And as the situation developed, it became increasingly difficult to separate substance from noise.

Genuine breaches, unverified claims, and recycled public data all began to coexist within the same narrative stream, each feeding into a growing sense that something larger and more coordinated was unfolding behind the scenes.

The first major breach emerged on March 20, when Handala claimed access to the email account of Deborah Oppenheimer, a former senior official involved in Israel’s intelligence and national security circles.

They claimed to have extracted and released more than 100,000 sensitive emails. The group framed it as exposing "Zionist propaganda, media manipulation, and covert operations" designed to promote Israeli interests globally. Israeli media treated the breach as genuine, noting it was part of a broader pattern targeting senior officials’ personal and work accounts.

Shortly afterward, between March 25 and 30, Handala escalated its campaign by announcing a breach of the personal Gmail account of Tamir Pardo, the former director of Mossad.

The group released a 14GB of documents and emails as "proof of concept." The material included sensitive personal correspondence, residential addresses, phone numbers (his and his assistant's), travel patterns, and internal documents. Among the highlights: a draft letter criticizing Israeli Prime Minister Benjamin Netanyahu and references to past operations, including alleged links to the assassination of Iranian nuclear scientists. Haaretz and other outlets confirmed the hack had occurred, describing it as part of Handala’s "hack-and-leak" psychological warfare campaign. The group even leaked some of Pardo's business-related emails and a letter addressed to the CIA chief.

On April 9, the campaign reached a new level with claims of prolonged access to a personal device belonging to Herzi Halevi, the former Chief of Staff of the Israel Defense Forces.

Handala said they had been "silently and relentlessly… right at the heart of General Herzi Halevi’s system" for years, collecting more than 19,000 confidential images and videos. What they published include footage and photos from sensitive military briefings, crisis rooms, and base visits; meetings with foreign officials that had been kept secret (e.g., with Jordan’s military chief and a previously undisclosed trip to Qatar involving U.S. Central Command leadership); strategic maps, classified documents, and clear (unblurred) faces of Israeli pilots, commanders, and security operatives; and personal and family moments mixed in with the professional material.

Taken together, these incidents reflect a deliberate strategy focused less on immediate operational disruption and more on psychological impact.

By exposing personal communications, internal dynamics, and glimpses of sensitive environments, the leaks aimed to undermine confidence, create reputational damage, and offer adversaries insight into the thinking and behavior of senior figures. This form of asymmetric cyber warfare prioritizes perception and long-term intelligence value over short-term tactical gain.

Against this backdrop, a separate post began circulating on April 10, claiming to reveal a comprehensive set of coordinates tied to Israeli intelligence sites, military facilities, and even private residences of political leaders.

The timing, coming just after a fragile ceasefire, gave the impression of a dramatic escalation and suggested that deeply classified information had been exposed.

However, closer examination revealed a far less alarming reality.

Unlike the earlier leaks, the coordinates post contained no supporting files, no internal documents, and no verifiable indicators of a breach. Instead, it presented a list of latitude and longitude points paired with brief descriptions—data that, in nearly every case, matched information already available through public sources such as satellite imagery, academic research, media reporting, and open-source intelligence platforms.

Locations such as Mossad’s headquarters in the Glilot area, major airbases, and even the Dimona nuclear facility have long been documented and discussed in public domains. Similarly, the listed residences associated with Benjamin Netanyahu corresponded to addresses that have been widely reported and photographed over the years.

None of this information required unauthorized access to secure systems; it could be compiled through basic research using widely accessible tools.

In the shadowy corners of the internet, some people try to post whatever they can: whether to help Iran, undermine Israel, embarrass the U.S., or simply chase attention in a moment of global tension.

What emerges from that chaos is not a single, coherent narrative, but a flood of competing claims, half-truths, and carefully packaged leaks. Some of it is real, backed by stolen data and quietly acknowledged by officials. But much of it is noise: recycled information, exaggerated claims, or entirely fabricated posts designed to ride the fear and curiosity surrounding the conflict.

This is not accidental.

Groups like Handala have helped blur the line between cyberattack and influence campaign, combining real intrusions with public messaging to amplify their impact. Analysts note that these operations are often designed as much for psychological effect as for technical disruption, turning even limited breaches into global headlines.

Israeli Prime Minister Benjamin Netanyahu and U.S. President Donald Trump: when two men has a different goal.

As a result, the online space becomes an extension of the battlefield itself. Information is weaponized, credibility is stretched, and timing becomes everything.

A single post, true or false, can spread rapidly if it aligns with what people already fear or expect to be happening behind closed doors.

In that environment, the most dangerous element isn't always the breach itself, but how things spread, and to whose hand the information fall.

Read: While Missiles Fly, Iran Rewrites The Rules Of Money Through Strait Of Hormuz, China, And Stablecoins

From Chill Stream To Golden Stream: How One Unfiltered Moment Turned A Small Streamer Into A Viral Phenomenon

Bess — Sat, 11 Apr 2026 11:04:24 +0000

10/04/2026

U-chan, the 21-year-old Japanese office lady known online as uchanniconico, has carved out a devoted niche on NicoNico with her completely unscripted facecam streams that feel less like performances and more like late-night conversations with a close friend who happens to be broadcasting her everyday life to thousands.

At level 78 on the platform, she regularly draws in viewers for relaxed sessions filled with casual chatter, light gaming, and the kind of raw, unpolished moments that bigger streaming sites would never allow, all while juggling her day job.

Her appeal lies in that total lack of filter: whether she's discussing random daily gripes, reacting to viewer gifts, or simply existing on camera with zero production value, and that authenticity has earned her a tight-knit fanbase that celebrates the messy humanity she brings to every broadcast.

On April 10, 2026, she kicked off another one of her signature drinking streams titled simply 酒雑談, starting around 8:50 PM JST and stretching for a full two hours and thirty-three minutes of sake-fueled miscellaneous talk that started off as the usual hazy, tipsy vibe her regulars love.

What began as a laid-back evening of sipping and bantering with chat quickly escalated when the alcohol worked its way through her system, hitting her bladder with an intensity that turned the stream into something far more memorable.

Viewers watching her live could see the shift in real time as U-chan's usual calm demeanor cracked under the mounting pressure of what she openly described on mic as reaching the absolute limit of a very human physiological need.

Instead of doing what any conventional streamer would: hitting pause, muting the mic, or simply ending the broadcast, U-chan sat there debating out loud in that unfiltered way she's famous for, weighing the choice between "配信継続" (keeping the stream going for her audience) and "解放" (finally giving in to relief).

The internal conflict played out right on her close-up facecam, her expressions cycling through strained smiles, fidgeting discomfort, and eventual reluctant acceptance of the only option that let her stay connected to the chat without interruption.

In a move that shocked even her most dedicated followers, U-chan made the call to continue rolling the camera as she headed straight for the toilet, choosing to handle the urgent bodily function without cutting away or hiding it from the live audience.

The footage captured every unedited second of the desperation giving way to pure, overwhelming release: her face softening into those exaggerated, blissful expressions of relief: eyes half-lidded, cheeks flushed from the sake. For modesty (at least), she had muted her microphone, but visually, nothing was hidden, with the tension written all over her face as it finally broke.

It wasn't staged or performative; it was raw, accidental lewdness mixed with genuine human vulnerability, the kind of moment that feels both hilarious and strangely intimate in the context of a public stream.

The chat section, already primed from the buildup, absolutely lost it in real time, flooding the screen with poop and pee emojis, rapid-fire shouts of “放尿,” “イキ顔” (orgasm face), and endless “神回” spam declaring the episode legendary, turning what could have been an embarrassing private moment into collective internet comedy gold.

The full timeshift recording of the entire broadcast lets anyone replay the full context: the relaxed start, the gradual buildup of discomfort during the sake chat, the pivotal decision point, and the aftermath where she returned to the stream still giggling and chatting as if nothing monumental had just unfolded.

Premium users or logged-in viewers can dive into the unedited two-plus hours to see how seamlessly she transitioned back into normal conversation, further cementing her image as someone who treats the audience like trusted confidants rather than distant spectators.

That single post exploded overnight, crossing one million views within hours and racking up thousands of likes, reposts, and quotes as the clip spread through Japanese otaku circles and beyond, with viewers praising the sheer commitment to keeping the stream alive no matter what.

This wasn't entirely new territory for U-chan, either fans quickly pointed out her history of similar "切り忘れ" incidents, like the December 2025 clip where she forgot to mute or cut during a bathroom break and the audio of her relieving herself leaked through anyway, tagged openly on NicoNico with おしっこ and 尿.

Those earlier moments had already built her reputation for pushing boundaries in the most relatable way possible, but the April 10 drinking stream took it further because the debate happened live on camera, the alcohol intensified everything, and the resulting expressions were so unmistakably meme-worthy that the edited highlight became inescapable.

In the broader world of Japanese live streaming, where NicoNico has long been a haven for this exact brand of unfiltered chaos, far removed from the polished, heavily moderated experiences on Twitch or YouTube.

And this incident perfectly encapsulates why the platform's core audience keeps coming back.

It's not just entertainment; it's a front-row seat to real life in all its awkward, urgent, and unexpectedly funny glory, the kind of content that feels forbidden on bigger sites but thrives here because the community embraces the messiness instead of shaming it.

What ultimately propelled this particular video into viral stardom wasn't shock value alone but the perfect combination of U-chan's genuine personality, the high-stakes drinking context that made the desperation hit harder, and the way the entire sequence played out as pure, unscripted comedy that anyone who's ever been caught between obligation and biology could instantly relate to.

The clip's rapid spread through X and fan communities, complete with reposts, reaction videos, and endless memes dissecting her expressions frame by frame, turned a routine evening stream into an overnight phenomenon that fans are already calling one of her most iconic "神回" moments.

For newcomers stumbling across the edited highlight, it serves as the ultimate gateway into the unique world of content creation: a reminder that sometimes, the best streams aren't the ones that go perfectly but the ones that embrace imperfection with zero apologies.

In this case, U-chan's April 10 drinking session stands as classic proof that in the niche corners of the internet, bodily functions and all, can win the audience by leaving them laughing, stunned, or eagerly waiting for whatever unfiltered adventure she broadcasts next.

From Chatbots To Cashbots: How Pika's AI Selves Can Help Users Earn While They Sleep

Aditya — Sat, 11 Apr 2026 04:51:04 +0000

The large language models (LLMs) war is intensifying.

Since the arrival of OpenAI's ChatGPT that ignited the battle towards AI supremacy, a rush among tech companies to build ever-more-capable conversational AI systems. What started as a novelty quickly reshaped how people accessed information, wrote code, and handled routine tasks, prompting competitors to flood the market with their own models and interfaces.

Amid this frenzy, most attention stayed on text-based chatbots and image generators, and here, Pika Labs took a different path.

Originally focused on turning text and images into short videos, the company quietly expanded its scope to something more personal: AI Selves, persistent digital extensions of individual users that combine appearance, voice, memory, and the ability to act across platforms.

And now, Pika Kabs announced a new layer to this system: users who create an AI Self can now earn tokens, redeemable for cash, whenever someone else chats with the agent, or makes use of its specialized skills. or ask for their photos.

Give birth to yours at https://t.co/AdSb9JGw6u

— Pika (@pika_labs) April 10, 2026

The announcement arrived with a short animated video illustrating the shift: from agents that consume tokens to ones that generate them.

It showed stylized AI characters fielding requests, sharing files, and delivering results, with an earnings dashboard displaying accumulated credits ready for cash-out. The feature is currently limited to interactions inside the Pika platform, and the company invited feedback through its Discord to refine how the system works in practice.

This move reframes the economics of personal AI.

Until now, most consumer-facing agents have been cost centers. Tools that require ongoing payments for compute, storage, or usage credits. By tying earnings directly to engagement, Pika turns the agent into a potential asset that can generate income without the owner needing to be online or actively involved.

A user might build an AI Self trained on their expertise in social media strategy, video editing, or fashion advice; others could then pay to consult it, with a portion of those fees flowing back as tokens.

The platform handles the transactions internally, converting earned tokens to cash once a threshold is met.

The implications extend beyond individual creators.

If AI Selves become widespread, they could scale personal presence in ways that were previously impossible. One person could maintain simultaneous conversations across messaging apps, professional networks, and content platforms while the AI handles routine inquiries, generates tailored responses, or executes simple tasks using its built-in skills.

For freelancers or small businesses, this blurs the line between human labor and automated output, potentially creating passive revenue streams that supplement or even replace certain services.

This is all very new, so we’re still working out the kinks. We’d love to hear how you think we can make it as valuable as possible. Join our discord community to share your thoughts: https://t.co/xB4AlOMAEo

— Pika (@pika_labs) April 10, 2026

At the same time, it raises practical questions about transparency.

When using this kind of feature, who is speaking, the human or the AI? And about how platforms will verify the quality or authenticity of these digital agents.

The broader disruption lies in how value flows through digital interactions. A marketplace of specialized AI Selves could emerge, where users discover and pay for agents built by others rather than building everything themselves. This might accelerate the creator economy by lowering the barrier to offering services at scale, but it could also pressure traditional jobs that involve advice, consultation, or repetitive content creation.

Companies might experiment with licensing popular AI personas or integrating them into customer support. Regulators and users alike will need to consider issues of data ownership, consent for training on personal likeness, and what happens when these agents evolve beyond their original prompts.

But for now, Pika's update is a single step in one platform, yet it points to a future where AI no longer merely assists but actively participates in the economy on behalf of its creators.

Published:

11/04/2026

Owning Vs. Renting Intelligence: The Battle For Autonomy, And The Strategic Divide Between Open And Closed AI

Danny Krisnanto — Fri, 10 Apr 2026 12:11:45 +0000

The debate surrounding open-source AI centers on a core strategic decision for organizations, developers, and institutions: whether to build and maintain ownership over their artificial intelligence capabilities or to rely on renting them from a limited set of dominant providers.

Closed models, typically offered through polished APIs and managed services, have gained widespread traction because they deliver immediate usability, reliable performance, and minimal setup requirements. Teams can integrate advanced capabilities quickly, focusing on application logic rather than underlying infrastructure or optimization challenges.

This approach suits rapid prototyping, standardized workflows, and scenarios where convenience and vendor-managed updates outweigh other priorities.

Yet renting intelligence creates inherent dependencies that extend beyond simple access.

Organizations become tied to external decisions on pricing models, rate limits, context handling, data retention policies, and even model availability or deprecation schedules.

Providers naturally prioritize their own commercial objectives, which may shift over time and introduce unexpected disruptions such as cost escalations or alterations in behavior that require rework in dependent systems. Data sovereignty concerns arise as sensitive information often flows to third-party servers, raising compliance and privacy issues in regulated industries.

In contrast, open-weight and open-source AI models empower users to host, inspect, modify, and deploy systems on their own terms, whether in private data centers, cloud environments they control, or even on-device hardware. This ownership model mirrors the broader technology landscape, where control over core assets has historically driven long-term resilience and innovation.

The parallels to earlier open-source movements in software are instructive.

Just as the dominance of proprietary operating systems and browsers gave way to collaborative alternatives that fostered competition, interoperability, and user empowerment, open AI approaches challenge concentrated control in intelligence technologies.

Early software ecosystems demonstrated how shared codebases could accelerate collective progress while preventing single-vendor lock-in. In AI, similar dynamics are unfolding: community-driven contributions enable faster iteration on architectures, training techniques, and efficiency improvements, often drawing from diverse global expertise rather than isolated corporate teams.

One of the most compelling practical advantages of open approaches lies in performance and economics.

Leading open models routinely achieve the vast majority of the capabilities demonstrated by their closed counterparts upon release, with community refinements frequently narrowing any remaining gaps within weeks or months.

Inference costs can drop dramatically, often by 70% or more, especially when models run on owned or optimized infrastructure rather than metered cloud services.

This efficiency stems from techniques like quantization, distillation, and specialized serving frameworks that allow organizations to tailor resource usage to their exact needs. Smaller, specialized models have matured to the point where capable intelligence operates effectively on consumer-grade devices, edge hardware, or resource-constrained environments, unlocking applications that demand low latency or offline functionality.

Beyond cost and performance, openness delivers deeper structural benefits.

Transparency allows developers and auditors to examine model architectures, weights, and sometimes training processes, facilitating the identification and mitigation of biases, vulnerabilities, or unintended behaviors.

This inspectability supports stronger accountability in high-stakes domains such as healthcare, finance, or public services.

Customization becomes straightforward: teams can fine-tune models on proprietary datasets, adapt them to niche languages or industry-specific terminology, or integrate them into private workflows without negotiating access restrictions.

For enterprises, this means avoiding vendor lock-in and building AI systems that evolve in lockstep with internal data and strategic goals. Governments and institutions focused on digital sovereignty gain the ability to operate models entirely within national borders or air-gapped networks, safeguarding sensitive operations from external influence.

The open ecosystem further amplifies these strengths through collaboration and shared infrastructure.

Global communities contribute improvements to foundational components, ranging from efficient inference engines and orchestration tools to standardized formats for model portability and evaluation benchmarks. Platforms that host repositories, provide serving libraries, and support distributed training lower the barriers to experimentation and scaling.

Organizations can leverage modular stacks for everything from data preprocessing to production monitoring, drawing on battle-tested open frameworks rather than starting from scratch.

This collective effort not only accelerates innovation but also distributes the burden of maintenance and security hardening across a wider pool of contributors, often resulting in more robust long-term outcomes than isolated proprietary development.

Open AI also aligns particularly well with emerging deployment patterns such as edge computing.

In scenarios requiring real-time decision-making. autonomous systems, industrial monitoring, smart infrastructure, or remote healthcare, processing data locally reduces latency, bandwidth costs, and exposure of raw inputs.

Models can run inference directly on sensors, gateways, or local servers, preserving privacy and enabling operation in environments with intermittent connectivity.

Enterprises in manufacturing, logistics, or energy sectors benefit from predictive maintenance, quality control, or optimization loops that operate independently of cloud round-trips. For public-sector applications, edge-enabled open models support everything from traffic management in smart cities to secure, localized analysis of citizen services without relying on foreign-hosted providers.

Despite these advantages, adopting open-source AI demands a clear-eyed assessment of its requirements and trade-offs.

Implementation typically involves greater upfront investment in technical expertise, hardware or cloud resources, and integration efforts. Teams must handle model serving optimization, security patching, performance monitoring, and lifecycle management tasks that closed services often abstract away.

Documentation quality and tooling maturity can vary across projects, requiring organizations to invest in internal capabilities or partner with experienced providers. Security considerations include the need for rigorous vetting of downloaded components to guard against supply-chain risks, adversarial attacks, or data-poisoning attempts that may surface more readily in publicly accessible repositories.

Quality control relies on community scrutiny rather than centralized testing, which can lead to inconsistencies in reliability or hallucination rates until additional validation layers are applied.

Licensing and compliance add another layer of complexity.

Not all releases labeled "open" provide identical levels of access, some offer weights alone while others include full training code or datasets, and terms can range from permissive to more restrictive commercial-use clauses.

Enterprises operating in regulated sectors must evaluate alignment with data-protection laws, audit requirements, and ethical guidelines, sometimes necessitating supplementary governance frameworks. Smaller teams or those prioritizing speed-to-market may still find closed platforms more pragmatic for initial exploration, where polished interfaces and dedicated support reduce operational overhead.

The maturing open-source AI landscape continues to address many of these hurdles through improved developer experiences.

Advances in user-friendly interfaces, automated deployment pipelines, comprehensive monitoring suites, and pre-optimized model variants are steadily narrowing the usability gap.

As inference hardware becomes more accessible and best practices for secure, scalable operations disseminate more widely, self-hosted or customized AI grows increasingly viable as a default rather than a niche choice. This evolution supports a hybrid reality in which open and closed models coexist: organizations select the right tool for each context, combining the rapid innovation and cost efficiency of openness with the polished reliability of proprietary services where appropriate.

Ultimately, the trajectory of AI development hinges less on raw benchmark scores alone and more on questions of control, adaptability, and equitable access.

Decision-makers across industries face an ongoing evaluation: the trade-offs of dependency on rented intelligence versus the responsibilities and rewards of ownership.

By fostering transparency, collaboration, and sovereignty, open approaches hold the potential to democratize powerful capabilities while mitigating the risks of concentrated power.

As supporting tools and practices advance, the choice between owning and renting intelligence becomes not merely technical but foundational to long-term strategic autonomy, innovation velocity, and resilience in an AI-driven future. Organizations that thoughtfully invest in the open ecosystem position themselves to shape and benefit from the next wave of technological progress on their own terms.

Technology

Business

Comparison

Expanding Creative Control, Runway Adds Custom Voices So Characters Can 'Sound Exactly Like You Want Them To'

Irvan — Fri, 10 Apr 2026 09:06:49 +0000

In the fast-moving world of generative AI, what many call the LLM war has expanded far beyond text, now encompassing video, audio, and interactive experiences.

Companies race to deliver higher fidelity, better consistency, longer sequences, and more intuitive controls, each trying to carve out an edge in simulating realistic worlds from prompts alone. Big players like Kling, Google’s Veo, and others push boundaries in photorealism, native audio integration, or extended clip lengths, while some earlier entrants have scaled back or pivoted.

The result is a crowded field where raw capability often clashes with practical usability for creators who need tools that actually fit into storytelling workflows.

Runway has approached this space differently, prioritizing precision and creative control over sheer spectacle.

Rather than chasing the longest possible video or the flashiest benchmarks, the company has built a suite of interconnected features aimed at filmmakers, animators, and narrative builders. Its models excel at maintaining character consistency across shots, offering fine-grained camera directives, and allowing iterative editing that feels closer to traditional post-production than pure generation.

This focus has kept Runway relevant even as the broader market fragments, with users frequently citing its reliability for professional-grade output that can be stitched into longer projects without losing coherence.

Custom voices are now available for Runway Characters. Design new voices from text prompts and start building Characters that sound exactly like you want them to. pic.twitter.com/WfhS38JXPD

— Runway (@runwayml) April 8, 2026

The latest addition is custom voices for Runway Characters, which fits squarely into that philosophy.

Characters themselves are a relatively recent capability, letting users create persistent, conversational video agents with defined appearances, personalities, and behaviors that can respond in real time or be directed through scripts. Until now, giving those characters a voice often meant relying on preset options, recording audio separately, or syncing external clips.

The new feature changes that by letting anyone design a voice directly through text prompts, much like prompting for visuals or motion.

Users can just describe what they want, like "a gruff, battle-worn male voice with a deep, raspy texture and a thick Highland brogue" or "high-pitched, low-confidence, clumsy viking voice," and let the system generates it, complete with natural intonation, accent, and emotional range.

The integration is seamless.

Once created, the voice can be assigned to a Character, and lip synchronization happens automatically when the agent speaks, whether in a generated video clip or a real-time interaction.

A single Viking warrior, for example, can shift from friendly and warm to aggressive and smoky or nervously comedic simply by swapping the voice prompt, all while the facial animation and body language remain consistent with the underlying model. This prompt-driven approach mirrors the rest of Runway’s workflow.

The idea of having this feature is to reduce friction by keeping everything inside one environment, eliminates the need for separate voice actors or heavy post-production, and gives creators the same level of textual precision they already expect for visuals.

Available to try now on the web app and to start building via the Runway API.

Learn more: https://t.co/8O71nDH0OW

— Runway (@runwayml) April 8, 2026

What stands out is how this update reinforces Runway's emphasis on usability within the larger competitive fray.

While some competitors focus on embedding audio natively into video generation or scaling to 4K with spatial sound, Runway has leaned into modularity and creator-centric tools, with Characters that feel like digital actors users actually can direct, edit, and evolve.

It's worth noting that the Runway Characters' custom-voice capability is far from being the most advanced audio model on the market. But at this time at least, it solves a specific pain point for people building stories, allowing them to prototype dialogue, test character dynamics, and iterate quickly without leaving the platform.

In a landscape where the next breakthrough is always one release away, Runway's steady accumulation of practical features suggests a different kind of strategy: one that values workflow depth over headline-grabbing specs alone.

As the war becomes fiercer, this kind of focused refinement may prove more enduring for those actually producing work than the raw power of any single model.

Published:

09/04/2026