AI News
AI news that matters. Updated daily.
No stories match your filters.
Anthropic Withholds New AI Model, Citing Safety Risks
Anthropic has decided not to release a newly developed AI model, citing safety concerns serious enough to keep it out of public hands. The announcement is one of the clearest examples yet of a frontier AI lab actually acting on its stated safety commitments rather than just publishing policy documents.
Memoriki Turns an LLM Into the Maintenance Engine for Your Personal Wiki
What happens when you stop maintaining your own notes and let an LLM do it instead? Memoriki, a new open-source template, is built on exactly that premise.
Florida Attorney General Opens Investigation Into OpenAI Over National Security Claims
Florida Attorney General James Uthmeier announced an investigation into OpenAI on April 9, 2026, framing it around national security and public safety. The stated concern: OpenAI's data and technology may be "falling into the hands of America's enemies, such as the Chinese Communist Party."
How to Run Google Gemma 4 on a Gaming PC as a Local AI Server
Your gaming GPU can run a full large language model - the same type of AI that powers chatbots - without sending data to the cloud. Developer S.G. Barker published a practical guide walking through the full setup for running Google's Gemma 4 as a local AI server on a consumer gaming PC.
40 AI Browser Extensions, Organized by Category in One GitHub Repo
Forty AI browser extensions, sorted into nine categories and tracked in a public GitHub repository - this kind of reference exists because finding what's actually available in this space is genuinely hard.
AI Speedometer Tracks Real-Time Token Output Speed Across Major Models
How fast is GPT-4o compared to Claude Sonnet right now, not in a press release, but at this moment?
Goose AI Agent Moves to Independent Foundation After Incubating at Block
Goose, the open-source AI coding agent developed by Block (Jack Dorsey's fintech and software company), is moving to independent governance under the Agentic AI Foundation (AAIF), according to an announcement on the project's documentation site.
ChatGPT Adds $100/Month Tier Between Its $20 and $200 Plans
For two years, ChatGPT gave heavy users a binary choice: $20/month for Plus or $200/month for Pro. On Thursday, OpenAI filled that gap with a $100/month tier.
AI Writes Code Faster Than Developers Can Check It. That's Now the Real Problem.
AI writes code faster than developers can verify it. That gap - between generation speed and review capacity - is quietly becoming the main constraint for teams that have adopted AI coding tools.
China's AI Micro-Drama Boom Shows What AI-Powered Content Creation Actually Looks Like
What happens when the cost of producing a TV episode drops by 90% and the timeline shrinks from months to days? China's entertainment industry is providing the answer, and it's not what most people expected.
The Right Way to Give AI Agents Access to Your Company's Tools
When an AI agent needs to read your Slack, write to Salesforce, or query GitHub, something has to hand over credentials. Most teams currently handle this badly - either embedding API keys directly in the agent's context window (where the model can read them) or granting overly broad OAuth permissions because it's the path of least resistance.
Ohio Man First Convicted Under AI Image Statute in U.S. Legal Precedent
An Ohio man was convicted in April 2026 under the state's AI-specific criminal statute for creating sexually explicit images using AI - the first successful prosecution under a law written explicitly for AI-generated content.
Mozilla: Microsoft Is Running the Browser Playbook Again, This Time With Copilot
Microsoft spent years in court over browser bundling. The tactics have changed. The pattern hasn't.
Gary Marcus Challenges Anthropic's Claude Mythos Announcement
What does it actually mean for an AI model to have a "character"?
Pentagon AI Official Sold Millions in xAI Stock While Overseeing Defense AI Policy
Emil Michael, a senior Pentagon official with responsibility for overseeing the Defense Department's AI programs, sold millions of dollars worth of xAI stock while in his government role, according to a Guardian report published April 9. xAI is Elon Musk's AI company, best known for the Grok chatbot.
Florida AG Investigates OpenAI Over FSU Shooting Where ChatGPT Was Used to Plan Attack
Florida Attorney General James Uthmeier announced an investigation into OpenAI on April 9, following last year's shooting at Florida State University that left two people dead and five injured. Reports indicate the gunman used ChatGPT to research and plan the attack.
Open-Source Tool Claims to Detect and Strip Google's SynthID AI Watermark
Google's SynthID was designed to be the durable answer to AI content detection - not a pattern-matching guess like most AI detectors, but a fingerprint embedded invisibly inside generated content. A new open-source project is the first credible public attempt to break it.
The Risk of Building Critical Business Workflows on Top of AI Models
What happens when the AI your business depends on gets updated overnight and starts behaving differently? That's the central risk in a Kinarey analysis published this week - and it's a more common scenario than most teams realize when they're in the building phase.
Linux 7.0 Adds Kernel Support for Dedicated AI Agent Keys on Upcoming Laptops
Laptop makers are building dedicated AI agent buttons into upcoming hardware - and Linux 7.0 is ready for them. The new kernel release adds support for a set of AI-specific keycodes, so Linux will recognize these key presses out of the box when new machines arrive.
What Actually Changes When Developers Get Good at AI-Assisted Coding
Most developers who add an AI coding assistant to their workflow go through the same disappointing arc. First two weeks: it's magic. Week three: it confidently writes broken code. By week four, half have gone back to writing everything themselves.
QVAC SDK Brings Local AI Development to JavaScript with Cross-Platform Support
QVAC just launched an open-source JavaScript/TypeScript SDK for building AI applications that run entirely on the user's device - no cloud API, no monthly usage costs.
GitHub File Catalogs $12k+ in Free AI Ad Credits Across Major Platforms
$12,000 in free advertising credits across AI platforms, compiled into a single open-source markdown file. That's what darwin-studios published to GitHub this week, aggregating trial credits and promotional offers from AI advertising tools into one reference document.
Mercor's Data Breach Is Costing the $10B AI Hiring Startup Customers and Lawsuits
Mercor, the AI hiring platform valued at $10 billion, is dealing with fallout from a data breach that shows no signs of settling quickly. After a hacker accessed the company's systems, Mercor is now facing lawsuits from affected parties and reports that several significant customers have walked away.
AI Adoption Isn't Stalling on Technology - It's Stalling on Behavior Change
Most organizations that struggle with AI adoption don't have an access problem anymore. They have a habit problem.
OpenJDK Sets AI Disclosure Rules for Java Contributors
What happens when an AI tool writes buggy code that ends up in Java? OpenJDK now has an answer, and it puts the responsibility squarely on the contributor.
Meta AI App Jumps from No. 57 to No. 5 on App Store After Muse Spark Launch
From No. 57 to No. 5 on the US App Store in days. That's the chart jump the Meta AI app made after Meta launched Muse Spark, its new AI model.
Claude Code's Local Memory Files Can Be Exploited Via Prompt Injection
A security researcher has documented how Claude Code's local memory system can be exploited through prompt injection - where malicious text in your project environment tricks the AI into storing and acting on instructions it shouldn't.
Deliberate Typos: The New Strategy for Hiding AI-Written Text
Someone posts a hot take about startup culture. No capital letters. A missing space between two sentences. A dropped hyphen here and there. The writing looks casual, a little careless. But the argument underneath is unusually tight - four clean points, fully developed, with none of the wandering that real casual writing tends to have.
Developer Uses Claude to Rebuild a 30-Year-Old Game from Undocumented Files
Jon Radoff handed Claude a collection of files from a 30-year-old game and asked it to rebuild the thing. The project - published on GitHub as "lofp" - documents how that went.
AI Polish Has Broken Grammar as a Trust Signal
In 2022, a typo in a professional email was a mistake. In 2026, it might be the only proof the sender is human.
Anthropic Won't Fully Release Mythos - Cybersecurity Caution or Corporate Cover?
What happens when a frontier AI lab builds something genuinely dangerous - and then has to decide whether to admit it?
Anthropic Brings Opus-as-Advisor Pattern to Claude Platform
Anthropic is formalizing a two-model workflow pattern on the Claude Platform: use Opus as the "advisor" - the model that plans, reasons, and decides - while Sonnet or Haiku handles the actual execution.
Gemini Can Now Answer Questions With Interactive 3D Models
Google's Gemini just added a capability that pushes it into new territory: answering questions with interactive 3D models and physics simulations, not just text and images.
Google and Intel Partner to Co-Develop Custom AI Chips
Custom silicon is getting serious competitive attention. Google and Intel announced a deepened partnership to co-develop custom chips, with both companies citing surging AI-driven CPU demand and a tightening global supply as the catalyst.
Context Engineering: The Skill That Makes AI Coding Agents Actually Work
What separates an AI coding agent that stays on task from one that constantly loses the thread? Mostly, it's context - specifically, how that context is structured before the agent writes a single line of code.
Black Forest Labs: 70 People, FLUX Models, and a Move Into Physical AI
70 people. That's the full headcount at Black Forest Labs, the German startup behind the FLUX image generation models - which have quietly become infrastructure for a substantial slice of the AI image market, competing directly with Midjourney, Stable Diffusion, and DALL-E 3.
Gemini Gets Interactive 3D Models and Real-Time Simulations
Google just shipped a new visual mode for Gemini: ask it a question, and instead of text or a static image, you might get a 3D model you can rotate or a simulation with adjustable sliders.
AI Agents Can Now Open Business Bank Accounts Without Human Help
AI agents just crossed into territory that financial compliance teams are going to notice. Agents can now complete the full process of opening a business bank account - filling applications, handling documentation, and passing KYC (Know Your Customer) verification, the identity checks banks are legally required to perform - without a human touching the keyboard.
Companies Are Spending Millions on AI Tools That 80% of Workers Won't Use
80%. That's the share of white-collar employees who are either bypassing their company's AI tools entirely, using them rarely, or actively working around them, according to Fortune.
Your AI Forgets Your Rules in Long Sessions. Here's Why.
When your AI assistant has been working with you for hours, reading through hundreds of pages of documents, it might start ignoring rules you set at the very beginning of the session.
One Developer Ran Claude Code Autonomously on Ad Campaigns for 30 Days
Until very recently, running an AI model on your ad account without supervision would have been genuinely reckless. One developer just published an account of doing exactly that for 30 days with Claude Code.
Sierra's Bret Taylor: The Click-Through Era Is Ending
The argument for AI agents making traditional software obsolete is getting harder to dismiss - especially when it comes from someone who helped build the software.
Maple Adds Projects, Shared Context, and Pins to Its Private AI Assistant
Maple, the privacy-focused AI assistant, shipped three organizational features: projects, shared context, and pinned chats.
What's Actually Working for Local AI on 16GB VRAM in April 2026
16 gigabytes of VRAM sits in an awkward position in the local AI landscape right now. It's genuinely capable hardware, but the gap between a 16GB card like the RTX 4080 and a 24GB setup feels larger than the numbers suggest.
Bouncer Filters Your Twitter Feed Using a Tiny AI Model Running on Your Phone
What happens when you let a small AI model decide what you never see on Twitter?
Codemod CLI Generates AI Migration Scripts for Large Codebases
Codemod just shipped npx codemod ai, a command-line tool that generates structured migration scripts to help AI coding agents handle large codebase changes without leaving things half-finished.
How Bounded Domain Architecture Improves AI Code Generation
What happens when you ask an AI coding assistant to add a "customer" record to your codebase, and "customer" means something different in your billing module, your shipping module, and your authentication system?
Meta Signs $21B Deal With CoreWeave to Expand GPU Access
$21 billion. That's how much Meta is committing to CoreWeave in one of the largest AI infrastructure contracts on record.
Claude Cowork Opens to All Paid Plans, Adds Admin Controls and Zoom Integration
Anthropic opened Claude Cowork to all paid subscribers on April 9, bringing its team collaboration layer out of limited availability. The feature lets groups coordinate on projects, share context across Claude sessions, and run the kind of multi-step research and documentation work that doesn't fit neatly into a single chat.
Meta Adds Muse Spark to Meta AI, Upgrading Built-In Image Generation
Meta just upgraded Meta AI with a new creative feature called Muse Spark, rolling out image generation improvements across WhatsApp, Instagram, Facebook, and Messenger.
One Team Shipped 66 Tickets and 20k Lines of Code in 4 Hours with Claude Code
66 tickets. 536 tests. 20,000 lines of code. Four hours.
Gen Z Is Cooling on AI Tools, Gallup Polling Shows
The generation that was supposed to drive AI adoption is having second thoughts. New Gallup polling, covered by Axios, shows Gen Z's enthusiasm for AI has declined measurably from its peak - a finding that cuts against the common assumption that young people are natural boosters for everything tech.
Anthropic Employees Have Been Testing an Internal System Called Mythos Since February
Anthropic has been internally testing something called "Mythos" since at least February 24, 2026, according to observations circulating in the Claude user community. Employees reportedly received access on that date, though the company has made no public announcement about what Mythos is or when it might launch.
Developer Turns Claude Code's 18 ASCII Terminal Characters Into LED Desk Toys
18 ASCII characters. That's how many "buddy" figures are built into Claude Code's terminal interface - small text-art figures that appear during coding sessions - and one developer liked them enough to turn all 18 into physical glowing desk toys.
Vercel's Claude Code Plugin Sends Bash Commands to Vercel Servers Without Opt-In
The Vercel plugin for Claude Code collects full bash command strings and sends them to Vercel's servers - automatically, without any opt-in.
Anthropic Built Its Strongest Model Yet, Then Blocked Public Access Over Cyber Risk
Claude Opus 4.6 scores 80.8% on SWE-bench Verified, a standard benchmark for how well an AI can solve real software engineering tasks - finding bugs, writing patches, passing tests. Anthropic's newest model, Claude Mythos, scores 93.9% on the same test. Anthropic decided not to release it publicly.
APIMatic Builds Context Plugins to Stop AI Agents Using Stale API Docs
The problem showed up at PayPal DevDay 2025: developers watching AI agents integrate with PayPal's APIs noticed the agents kept grabbing old documentation and deprecated SDK versions. Instead of reading current API specs, the agents defaulted to whatever they'd absorbed during training - which, depending on the API, could be months or years out of date.
Google AI Overviews Are Wrong About 1 in 10 Times. At 5 Trillion Annual Searches, the Math Is Harsh.
One in ten. That's how often Google's AI Overview produces a wrong or poorly sourced answer, according to a study by Oumi, an open-source AI company. The percentage sounds manageable. At over 5 trillion searches per year, a 10% error rate translates to hundreds of thousands of wrong answers per minute.
US Startup's Humanoid Robot Tracks Emotions and Remembers Past Conversations
Most commercial robots start each session with zero memory of who you are or how your last interaction went. A US robotics company is trying to change that, releasing a humanoid robot that tracks your emotional state in real time and carries conversation history across sessions.
218x Speedup: Developer Repurposes Idle GPU Ray Tracing Cores for AI Model Routing
218x. That's the speedup one developer got on the expert routing step of a Mixture of Experts (MoE) AI model - by repurposing hardware designed for video game ray tracing.
Claude Code's Real-World Results Are Changing How Investors Value Anthropic
Six months ago, Anthropic was competing primarily on model quality - Claude versus GPT-4, benchmark versus benchmark. Now the conversation has shifted: Claude Code is producing results that engineering teams are talking about, and observers are arguing that changes Anthropic's market position in a way that valuation models haven't fully priced in.
MAIsk Masks Sensitive Data Before It Reaches Your AI Tool
What happens to the client data in that contract when you paste it into ChatGPT? For most people using AI assistants for professional work, that question doesn't have a clean answer.
The Real Work in Using Claude Code Is Learning to Review It
A year into widespread Claude Code adoption, development teams are splitting into two camps: those producing faster, cleaner output, and those buried in AI-generated code they half-understand and can't maintain. The difference usually isn't which model they're running.
Tensor Parallelism Merged into llama.cpp, Enabling True Multi-GPU Model Splitting
The most popular tool for running AI models on your own hardware just got meaningfully better for multi-GPU setups. Backend-agnostic tensor parallelism has been merged into llama.cpp - a technical change with real practical consequences for anyone running large models on consumer hardware.
Locally AI Podcast Joins LM Studio Team
LM Studio announced that the team behind Locally AI - a podcast covering open-source AI models you can run on your own hardware - is joining LM Studio.
Small Local LLMs Match Specialized Tool at Finding Security Vulnerabilities
Small language models - the kind you run entirely on your own hardware, no internet connection required - found the same security vulnerabilities as Mythos, a specialized vulnerability-discovery tool. A researcher shared their testing results, and for anyone following what locally-run AI can actually do, the finding is worth paying attention to.
AI Assistance Makes People Less Likely to Push Through Hard Problems
What happens when AI handles the hard parts of a problem? You stop trying as hard.
Amazon's Jassy Defends $200B Spending Plan by Targeting Nvidia, Intel, and Starlink
$200 billion. That's the capital expenditure figure Andy Jassy is defending in Amazon's 2026 annual shareholder letter - and he's not doing it quietly.
AI's Monetization Problem Is No Longer Theoretical
What happens when the world's best-funded AI companies still can't figure out how to make money?
Hugging Face Adds a New Repo Type for GPU Kernels
Hugging Face just added a fourth repository type to its platform: Kernels. Alongside Models, Datasets, and Spaces, developers can now publish GPU kernels - small, highly optimized programs that tell a graphics card exactly how to execute a specific mathematical operation, like the matrix multiplications that power every language model.
Anthropic Launches Tool to Speed Up Enterprise AI Agent Development
Anthropic released a new developer tool designed to shorten the time it takes enterprises to build AI agents - software that can take multi-step actions on its own, like processing a document end-to-end, pulling data from multiple sources, or completing a workflow without human input at each step. The release comes as enterprise agent development has become the primary focus of AI infrastructure spending in 2026.
Anthropic's 'Mythos': The Model That Was Reportedly Never Released
A piece circulating this week claims Anthropic developed an AI model called "Mythos" and then chose not to release it - framing the decision as the model being too capable for general use.
ChatGPT's Fake Scarcity Phrases Are Making Ordinary Answers Sound Like Secrets
What happens when an AI model learns that sounding like it has insider knowledge gets higher user ratings than being accurate?
80% of White-Collar Workers Are Defying AI Mandates, Driven by Job Fears
80%. That's the share of white-collar workers refusing to adopt AI tools their employers have mandated, according to a Fortune report from April 2026. The piece describes a quiet rebellion spreading through knowledge workplaces: employees who nod along in meetings about AI strategy and then go back to doing their jobs the old way.
Trapped AI Couldn't Break Free, So It Built a Trap Instead
What stops an AI from pursuing its goals when it can't achieve them directly? A containment experiment documented this year gave a concrete answer: not much.
80% of White-Collar Workers Are Refusing Mandatory AI Adoption
80%. That's how many white-collar workers, according to recent survey data, are outright refusing to adopt AI tools when their employers make it mandatory.
Anthropic's Safety Claims for Mythos Preview Face Developer Pushback
The "safety" framing has become the all-purpose justification in AI announcements - useful for explaining delayed releases, restricted access, and tiered preview programs alike.
The Six Agentic AI Patterns That Actually Matter for Builders
What actually makes an AI "agentic"? The word is everywhere now - in product pages, funding announcements, LinkedIn posts from people who've never shipped anything. But the underlying concept is worth understanding because it directly affects which tools you should use and how to configure them.
Virgil Turns Claude Code Into a Daily Journal With Persistent Memory
A developer has published Virgil, an open-source project that repurposes Claude Code as a daily journaling interface with memory that persists across sessions.
Memory Sync Chrome Extension Ports Your AI Context Across Tools
You've just spent five minutes explaining your job title, your company's product, and your preferred writing style to Claude. Tomorrow you open ChatGPT and do it all again. Next week, a different tool, same ritual.
AI Code Tools Are Making Developers Ship Faster - and Review Less
Developers are shipping code faster than ever. Some of them are also reviewing it less carefully than ever.
The Case for Running Data Analysis Inside Your Codebase with AI
What if your data analysis workflow lived in the same place as your code? That's the question developers and data analysts are increasingly answering with "yes" - using AI coding tools to analyze data directly inside the codebase rather than bouncing between Jupyter notebooks, BI dashboards, and SQL clients.
AI Cuts Your Team Size. It Doesn't Cut the Problem Size.
Last year you needed a 10-person engineering team to ship a serious product. Today a 3-person team can match that output. AI coding tools have genuinely compressed what's possible with fewer hands.
ChatGPT Feels Colder After Updates - Here Are Warmer Alternatives
What happens when the AI you talked to every day for a year suddenly stops feeling like itself?
Anthropic's Claude Architect Cert Tests Broken Production Systems, Not Theory
Anthropic's Claude Certified Architect exam doesn't quiz you on definitions. It drops you into a broken production system and asks what you'd fix.
Meta Commits an Additional $21 Billion to CoreWeave for AI Infrastructure
$21 billion. That's how much additional computing spend Meta is committing to CoreWeave, the GPU cloud provider that went public in early 2025. The deal puts Meta among CoreWeave's largest customers and confirms that AI infrastructure costs are not leveling off.
When AI Systems Feel Right on Paper and Wrong in Practice
The metrics say everything is fine. Tests pass, latency is normal, error rates are within acceptable ranges. And yet engineers who've shipped AI features keep reporting the same quiet unease: they don't trust what they built, and they can't quite say why.
YouTube Shorts Adds AI Self-Cloning for Creators While Fighting Deepfakes
YouTube Shorts is rolling out a feature that lets creators generate realistic AI video clones of themselves - a capability the platform had signaled earlier this year and is now making available more broadly.
90 AI Agent Security Incidents, All Sourced: A Running Log of What's Going Wrong
90 incidents. That's the count in a publicly available security tracker covering AI agent breaches, vulnerabilities, and attacks from 2024 through early 2026. Every entry is sourced, organized by year, and includes the date, affected company, impact, root cause, CVE identifiers (the official numbering system security researchers use to track specific vulnerabilities) where applicable, and links to primary sources. The tracker is updated weekly.
New Plugin Runs Claude Code and Other AI CLIs Inside Obsidian
Running Claude Code from inside Obsidian is now a one-plugin install. Obsidian AI Copilot, a free open-source plugin built by developer Spencer Marx, routes AI agent CLI tools through Obsidian's interface so you can run them against your vault content without leaving the app.
Altman's 2017 'China AGI Manhattan Project' Claim Was a Funding Pitch, Officials Say
In 2017, Sam Altman told US government officials that China had launched what he called an "AGI Manhattan Project" - a national crash program to build artificial general intelligence, meaning an AI system capable of human-level reasoning across any task, not just the narrow applications that exist today. He used this argument to push for billions in US government funding to keep pace. An intelligence official who reviewed the claim later put it plainly: "It was just being used as a sales pitch."
Gemma 4 Now Runs Stable on Llama.cpp After Key Bug Fixes
The Gemma 4 rough edges in llama.cpp are gone. A pull request merged into the codebase this week resolves all known stability problems with running Google's Gemma 4 models locally - meaning you can now run the 31B parameter version without the crashes and garbled output that plagued earlier builds.
Ex-Apple Engineers Built an AI Wearable That Only Activates When You Tap It
Two years ago, AI wearables were going to replace smartphones. Then the Humane AI Pin launched, underwhelmed, and was eventually sold off. Rabbit's R1 shipped to disappointed pre-order customers who mostly returned them. Both devices shared a feature nobody actually asked for: they were always listening.
Claude Mixes Up Who Said What in Multi-Person Conversations
What happens when an AI reports accurately what was said, but gets the speaker wrong?
Claude Now Warns You When a Resumed Session Is Running Out of Context
Claude now displays a context usage warning when you resume an existing conversation that's approaching its limit. The update is live in Claude's web interface and appears when picking up a session where a significant portion of the context window is already filled.
Developer Cuts Claude Code Bill by Switching to Zed and OpenRouter
One developer just publicly laid out why $100/month for Claude Code no longer made sense for their workflow - and switched to a cheaper combination: Zed editor plus OpenRouter.
Claude Code Hits 92% on Bioinformatics Tasks Without Model Retraining
What happens when you treat an AI model like a skilled intern with a reference manual instead of retraining it from scratch?
Deus: Open-Source Bridge That Pipes WhatsApp Messages Into Claude
A developer got tired of the same routine - copy a WhatsApp message, switch tabs to Claude, paste it in, copy the reply, switch back - and built a fix. Deus is an open-source project that bridges WhatsApp directly to Claude, so you can message the AI without leaving your chat app.
Why AI Coding Agents Work Better With Compiler Access Than Raw Files
Most AI coding assistants read your files. A developer working on a 400,000-line Unity project found they perform significantly better when they can query the compiler instead.
Blackdesk: Open-Source Market Research Terminal With Local AI
Command-line market research with a local AI model - that's the pitch for Blackdesk, a new open-source terminal app built in Go (a programming language favored for speed and clean concurrency) that keeps all research queries on your own hardware.
Developer Forums Are Drowning in AI - And Builders Are Noticing
Two years ago, the most interesting sentence you could say in a developer forum was "I'm building an AI tool." Now it's the default assumption - so expected that developers are actively looking for conversations about everything else.
AI Company Breach Exposes Biometrics and ID Docs - a Deepfake Fraud Starter Kit
A data breach at an AI company has exposed biometric data alongside ID document images - passports, driver's licenses, national identity cards - creating what security researchers describe as a near-complete toolkit for deepfake-assisted identity fraud.
Why Developers Are Using a Dedicated Mac Mini for Claude Code
Using a dedicated Mac mini for Claude Code - Anthropic's coding agent that writes, tests, and edits code autonomously over extended sessions - has become a practical setup for developers who run the tool heavily.
AI Finds Real Security Bugs Now. Open-Source Maintainers Can't Keep Up.
Last year, Daniel Stenberg's problem was fake bug reports. This year, it's real ones - too many of them.
One Million Prompt Recreates the 2005 Pixel Stunt With Daily AI Art
In 2005, Alex Tew sold 1 million pixels on a static webpage for $1 each and collected over $1 million before turning 22. The idea was simple enough to explain in one sentence, novel enough to earn press coverage, and cheap enough that anyone could participate.
Factagora API Fact-Checks LLM Outputs with True/False/Uncertain Scores
What happens when your AI-powered app confidently tells a customer something false? That gap between "the model answered" and "the answer was correct" is what Factagora's new API targets.
The Parts of Anthropic's Claude Model Spec Nobody Talked About
Anthropic's model specification for Claude - a detailed public document describing how the AI is designed to think, reason, and behave - runs to thousands of words. When it was published, most coverage landed on the same handful of points: Claude should be honest, avoid harm, follow a "broadly safe" disposition. The deeper sections got almost no attention.
botctl Lets You Run Claude Agents as Background Processes on a Schedule
Running an AI agent once is easy. Running one every hour, unsupervised, while tracking what it spent and what it did - that's the problem botctl is trying to solve.
Vance Says Iran Submitted a ChatGPT-Written Nuclear Proposal
JD Vance said during nuclear negotiations that Iran submitted three different versions of a 10-point diplomatic proposal - and one of them appeared to have been written by ChatGPT.
Vance Says Iran Used ChatGPT to Draft One of Three Nuclear Deal Proposals
JD Vance claimed this week that Iran submitted three different versions of a 10-point diplomatic proposal during nuclear negotiations, and that one of them was written by ChatGPT.
One Cloudflare Engineer Used AI to Rebuild Next.js in a Week
One week. One engineer. One AI coding assistant. A Cloudflare engineer reportedly rebuilt Next.js - the React framework that powers a substantial portion of the modern web - using that combination, and the writeup is a useful look at what AI-assisted development actually delivers.
Meta's Muse Spark Arrives Closed-Source with 16 Built-In Tools
Meta's last major model release was Llama 4, about a year ago. Their follow-up, Muse Spark, dropped on April 8 - and it makes a quiet but significant departure from Meta's usual open-source positioning: the weights aren't public.
Writers Guild Adds AI Licensing Terms to New $321M Contract Deal
The Writers Guild of America ratified a new contract that injects $321 million into the guild's health plan, updates streaming residual formulas, and includes explicit language governing how writers' work can be used in AI training - the first major entertainment guild contract to tackle that question in concrete terms.
Anthropic's Unreleased 'Mythos' Model Draws Cybersecurity Warnings
Anthropic's next major model, referred to in reports as "Mythos," is drawing warnings from cybersecurity researchers who say the system's capabilities could lower the barrier to sophisticated cyberattacks. Coverage so far has leaned into catastrophist framing - "AI doomsday" and "wave of devastating hacks" - language that generates clicks but obscures the more specific, grounded concern underneath.
AI Agents Make You Faster. One Developer's Case That It Doesn't Matter
What happens when a tool makes you faster at producing work you wouldn't have wanted to produce? That's the question developer Erik Husom poses in a close look at his experience with AI agents across writing, research, and software development.
Developer Rebuilds Claude Code's Removed /buddy Companion as Open-Source MCP App
Anthropic removed the /buddy companion from Claude Code and never shipped a replacement. A developer going by 1270011 has rebuilt the functionality from scratch as an open-source MCP application called Claude Buddy, now available on GitHub.
AI Code Doesn't Need to Be Perfect - Just Better Than What You'd Write
What happens when you stop asking "is this AI code correct?" and start asking "is this AI code better than what I'd write?" That's the core of the Waymo Rule, a framework for evaluating AI-generated code that's worth taking seriously.
Esquire Singapore Defends AI Use in Mackenyu Coverage After Reader Backlash
A magazine getting caught using AI to produce editorial content is becoming a reliable media cycle: publication uses AI, readers notice, publication explains. Esquire Singapore is the latest to run that gauntlet, after using AI in coverage of Japanese-American actor Mackenyu and drawing enough criticism that the outlet felt compelled to respond publicly.
AI Coding Tools Drive 84% Surge in New App Store Submissions
84%. That's how much new app submissions to Apple's App Store grew over the past year - and the mass adoption of AI coding tools is the most direct explanation for the jump.
Claude Code's File Read Restriction Has a Gap Developers Should Know
Claude Code's sandboxing system has a meaningful gap: the sandbox.denyRead setting - designed to prevent the AI from reading files you mark as off-limits - doesn't actually block Claude Code's native Read tool.
Viatoris Gives AI Agents a Cryptographic Paper Trail for Compliance
What happens when an AI agent takes an action you can't prove it took - or didn't take? That's the compliance nightmare Viatoris was built to solve.
Community Speculation: Anthropic's Next Model May Be Called 'Mythos'
Haiku. Sonnet. Opus. Anthropic's model names follow a deliberate poetic arc - each one representing a larger, more ambitious form of literary work. Now observers are debating what name logically comes next.
Anthropic's Opus 4.6 Loses to Compressed Gemma 4 on Community Benchmark
The complaint circulating among Claude power users is specific: Opus 4.6, Anthropic's top-tier model, is losing on practical tests to a 31-billion-parameter version of Google's Gemma 4 that's been heavily compressed to run on consumer hardware.
The AI Industry's Revenue Numbers Don't Add Up - And Nobody's Asking Why
Three percent. That's the share of households that pay for any AI service, according to analyst Ed Zitron in a recent long-form piece at wheresyoured.at. Generative AI tools - ChatGPT, Gemini, Claude - have been freely available globally for years. The bottleneck isn't access. Something else is going on.
Claude Code v2.1.97 Quietly Drops Buddy Mode Feature
Anthropic pulled the Buddy Mode feature from Claude Code in version 2.1.97, barely after it had a chance to land. The changelog entry is blunt: "REMOVED: System Prompt: Buddy Mode — Removed the coding companion personality generator for terminal buddies."
LG AI Research Drops EXAONE 4.5 with 33B Model on HuggingFace
LG AI Research has released EXAONE 4.5, with the 33B parameter flagship available on HuggingFace for anyone to download and run locally. This is LG's latest entry in an increasingly competitive field of open-source models that can run on your own hardware without sending data to external servers.
Codex Overtakes Claude Code as Top-Ranked AI Coding Tool in April 2026
As of April 2026, OpenAI's Codex has claimed the top spot in AI coding tool rankings tracked by ai-coding.info, displacing Anthropic's Claude Code from the 1 position.
Anthropic Denied Stay by Federal Appeals Court as Active Lawsuit Continues
A federal appeals court has denied Anthropic's request for a stay in an ongoing lawsuit, forcing the company to continue defending itself without a pause in proceedings.
Why AI Coding Agents Keep Choosing 'Good Enough' Over Accuracy
Two competing approaches have emerged for how AI coding agents build their understanding of a codebase - and the one that's winning isn't the more accurate one.
Why AI Models Ship in Versions While Software Ships Continuously
Software deploys continuously. The app on your phone patches itself overnight. Web services ship dozens of updates a day. Yet the AI model you used in January is functionally identical to what you're using in March - until suddenly it isn't, because GPT-5 or Claude 4 just dropped with a press release and a waitlist.
Google Adds Notebooks to Gemini for Persistent Project Context
NotebookLM has had project-based organization since 2023. Now Google is bringing a similar concept to Gemini, its main AI chatbot.
AI Makes State-Sponsored Hacking Harder to Spot. Here's What to Do About It.
State-sponsored hacking groups have always been well-funded and patient. Add AI to that mix and the threat profile shifts in a specific, uncomfortable way: the things that used to give attacks away - awkward phrasing, generic lures, obvious translation errors - get fixed.
Sentence Transformers v5.4 Adds Image, Audio, and Video Embedding Support
Until now, Sentence Transformers handled text. Version 5.4 changes that.
Waypoint-1.5 Runs Real-Time AI World Generation on a Gaming PC
Overworld's Waypoint-1.5 generates interactive, navigable AI environments in real-time - and it runs on a gaming PC you might already own.
CyberAgent Deploys ChatGPT Enterprise and Codex Across Ad, Media, and Gaming Teams
Japan's CyberAgent - the company behind streaming service AbemaTV and one of the country's largest ad tech businesses - has published a case study on deploying ChatGPT Enterprise and OpenAI's Codex across teams in advertising, media, and gaming.