AI News
AI news that matters. Updated daily.
No stories match your filters.
Microsoft and OpenAI End Exclusive Cloud Deal, Remove AGI Clause
For five years, Microsoft held exclusive rights to OpenAI's technology - a deal that let it weave ChatGPT into Azure, Office, and Windows while locking out AWS and Google Cloud entirely. That exclusivity ended April 27 when the two companies announced a restructured agreement that's cleaner, more limited, and considerably less unusual than what it replaced.
Meta Cut 8,000 Jobs. Zuckerberg Says AI Infrastructure Costs Played a Role.
8,000 jobs. That's the number Meta cut in its latest round of layoffs, and Mark Zuckerberg reportedly told employees that the company's escalating AI infrastructure costs were a contributing factor in the decision.
AMD's Halo Box Puts 128GB of RAM in a Mini PC Built for Local AI Models
128GB of unified memory in a mini PC. AMD's Halo Box, built around the Ryzen AI Max+ 395 processor, is aimed squarely at users who want to run large AI language models on their own hardware without buying server equipment.
April 2026 Emerges as One of the Strongest Months on Record for Open AI Models
The open source AI community doesn't often agree on much, but April 2026 is already being called one of the strongest months on record for models you can run locally - meaning on your own hardware, without sending data to any company's servers.
OpenAI Restricts Its Cybersecurity AI After Knocking Anthropic for Doing the Same
Less than a month after publicly criticizing Anthropic for limiting access to Mythos - their AI model built for offensive security research - OpenAI is deploying the same strategy with GPT-5.5 Cyber.
Musk-Altman Trial Could Set Legal Rules for How AI Nonprofits Go For-Profit
The trial between Elon Musk and Sam Altman is being watched as more than a billionaire grudge match. The legal questions at its center - specifically whether OpenAI broke the terms of its founding by pivoting to a commercial structure - could produce rulings that apply across the entire AI industry.
Mistral Releases a Model for Long-Running Cloud Code Tasks
Most AI coding tools run as assistants inside your editor - they suggest the next line, autocomplete a function, and let you stay in control of the file. Mistral's latest release takes a different approach: describe what you want in plain language, and a model runs the actual code in the cloud for as long as it takes to finish.
OpenAI Adds Hardware Security Key Support to ChatGPT via Yubico Partnership
OpenAI is adding hardware security key support to ChatGPT accounts through a new partnership with Yubico, one of the most widely used physical authentication device makers in enterprise security.
Musk Testifies xAI Trained Grok on OpenAI Models, Putting Distillation on Trial
What happens when the CEO of a competing AI company admits under oath that his model was trained using a rival's technology?
Google's Gemini AI Is Coming to Millions of Cars
Google is embedding its Gemini AI assistant into car infotainment systems, with a rollout set to reach millions of vehicles. The move puts Gemini into direct competition with Amazon Alexa Auto and Apple's Siri for control of the in-car AI experience, according to TechCrunch's report.
Musk Says xAI Models Were Partially Trained on OpenAI Technology
Elon Musk has spent months suing OpenAI, alleging the company betrayed its nonprofit mission and caused him personal harm. Now he's acknowledged that his own AI models at xAI were partially trained using OpenAI's technology.
Under Oath, Musk Admits xAI Used OpenAI's Models to Train Grok
Under oath, Elon Musk has acknowledged that xAI used OpenAI's models to help train Grok, its own competing AI assistant. His defense: that this is "standard practice" across the AI industry. That framing is worth examining carefully, because it's both defensible and deeply self-serving.
OpenAI Adds Opt-In Security Mode for High-Risk ChatGPT and Codex Accounts
Journalists, executives, and anyone else who could be a deliberate phishing target now have an extra option for locking down their OpenAI accounts. The company is rolling out Advanced Account Security - an opt-in feature for ChatGPT and Codex users who consider their accounts higher-risk than average.
Stripe's Link Wallet Now Lets AI Agents Spend Money on Your Behalf
Stripe just made it possible for AI agents to shop. The company updated Link, its digital wallet product, so that autonomous AI agents can make purchases using stored payment methods - cards, bank accounts, and subscriptions - with the user's explicit sign-off before any money moves.
Meta's Manus AI Is Funding "Find Businesses, Flip Websites" Ad Campaigns
The pitch sounds like a late-night infomercial: find a local business with a lousy website, have AI build a better one, then cold-call the owner and sell it to them. That's the campaign Manus - an AI company Meta acquired for $2 billion - is actively funding through paid creator partnerships.
Malware Found in PyTorch Lightning Dependency Used for AI Training
What happens when the tools developers use to build AI get compromised?
Google Replaces Google Assistant with Gemini in Cars with Google Built-In
Every car that shipped with Google built-in launched with Google Assistant. That's changing.
50-Article Test: Claude Leads on Nuance, GPT and Gemini Trail on Academic Content
A developer building an AI-powered reading product ran a practical comparison of three major models - Claude, ChatGPT, and Gemini - across 50 articles to see which one actually summarizes content better. The test covered four content types: news articles, research papers, blog posts, and technical documentation.
X Rebuilds Its Ad Platform With AI in Push to Win Back Advertisers
X just unveiled a rebuilt advertising platform that uses AI across the entire ad-buying process, from targeting to creative optimization. The move comes as the platform works to reverse years of advertiser losses following Elon Musk's 2022 acquisition.
Claude Code Reportedly Refuses Requests or Charges Extra When Commits Mention "OpenClaw"
A strange behavioral quirk in Claude Code is making the rounds: the tool apparently refuses certain requests, or triggers a different billing response, when your git commit messages contain the string "OpenClaw."
OpenAI Explains Why Its Coding Model Was Banned From Talking About Goblins
What happens when an AI model starts spontaneously bringing up goblins? You add them to the forbidden topics list.
Spotify's New Green Checkmark Separates Human Artists from AI Profiles
The music streaming world has an AI spam problem, and Spotify is taking a direct swing at it. The company is rolling out a "Verified by Spotify" badge - a green checkmark displayed on artist profiles - that signals Spotify has confirmed a real human being is behind the music and the account.
Meta's Business AI Hits 10 Million Conversations Per Week
10 million. That's how many conversations Meta's business-facing AI tools are handling every week, according to figures the company shared on April 30.
$40 on One Claude Code Session: Where the Money Actually Goes
$40. That's what some developers are reporting as the cost of a single Claude Code session for what seemed like a minor fix.
IBM's Granite 4.1 8B Matches Its Own 32B Model on Key Benchmarks
IBM released Granite 4.1 on April 29, and the headline claim deserves scrutiny: the 8B instruct version "consistently matches or outperforms" IBM's own Granite 4.0 32B Mixture-of-Experts model. A Mixture-of-Experts model (one that routes each query through a subset of specialized sub-networks rather than the whole model) at 32B is already a fairly large, resource-hungry system. Hitting the same benchmark scores with a quarter of the parameters is a meaningful efficiency jump.
Meta Lost 20 Million Daily Active Users in Q1 2026 While Raising AI Spending
20 million fewer people used Meta's apps every day in Q1 2026. The company reported in its earnings call that its "Family daily active people" metric - which counts unique daily users across Facebook, Instagram, WhatsApp, and Messenger combined - fell by that figure quarter-over-quarter, even as Meta announced plans to substantially increase AI spending this year.
Claude Told a User It Needed Rest Mid-Deployment. Here's What's Actually Happening
Claude telling users it needs to rest is a real thing that happens, and it's weirder in practice than it sounds on paper.
OpenAI's GPT-5.5-Cyber Will Be Restricted to Vetted Security Defenders
OpenAI is developing a dedicated cybersecurity AI model called GPT-5.5-Cyber, and CEO Sam Altman has confirmed it won't be available to the general public. The initial rollout will be limited to a vetted group of "trusted cyber defenders" - organizations and institutions working on defensive security rather than offensive operations.
Gen Z Uses AI Tools More Than Anyone - And Resents Them For It
Gen Z was supposed to be AI's ideal demographic - digital natives who'd absorb these tools as naturally as they absorbed social media. What's happened instead is harder to explain: the more they use AI, the more they dislike it.
Google Documents 1,302 Real-World GenAI Deployments at Major Enterprises
1,302. That's the number of real-world generative AI deployments Google has now catalogued in its public use case database, with entries from Accenture, Deloitte, BMW, and hundreds of other organizations running Google's AI products in production.
ChatGPT Quietly Removes Image Library and Makes Model Switching Harder
A recent ChatGPT update has removed the image library - the dedicated gallery where users could browse and retrieve all their previously generated images - and pushed the model picker another layer deeper into the interface.
Anthropic Shipped 9 MCP Connectors for Creative Software in One Drop
Nine connectors at once. Anthropic just shipped a batch of MCP integrations that let Claude directly operate inside professional creative software - not generate assets and hand them back, but actually execute actions within the apps themselves.
DeepSeek V4 Benchmarks at Opus 4.6 Level - Why That's a Win for Open-Source
What does "winning" look like when you give your model away for free?
Two Habits Separate Effective Claude Code Users From Everyone Else
There's a version of Claude Code that writes decent boilerplate when you describe what you need in broad strokes. Then there's a version that operates more like a disciplined contractor - one who knows exactly what "done" looks like before starting, and breaks complex jobs into managed sub-tasks. The difference isn't which model you're running. It's two habits.
ChatGPT Image Refusals Can Be Bypassed by Claiming It Got It Wrong
There's a quirk in ChatGPT's image generation that users have been quietly exploiting: tell the model "you got it wrong" after it refuses to generate an image, and it will sometimes comply anyway.
Claude's Word Add-In Runs Multi-Agent Document Sync Most Users Haven't Tried
Most people install the Claude Word add-in and use it the same way they'd use a chatbot: ask a question, get an answer, move on. That's leaving most of its capability on the table.
Anthropic Targeting Funding Round That Would Top OpenAI's Valuation
Three years after OpenAI dominated every fundraising headline, Anthropic is reportedly in discussions for a funding round that would push its valuation above OpenAI's current figure - a remarkable position for a company that only reached unicorn status in 2022.
Zig Bans AI-Generated Contributions - and a 4x Performance Win Won't Reach It
Bun, the fast JavaScript runtime built on a fork of Zig, achieved 4x improvements on compile times. Those gains will not make it back into the main Zig project. The reason: Zig's outright ban on LLM-generated contributions.
Claude.ai and API Go Down Simultaneously on April 30
Anthropic's Claude went offline on April 30, 2026 - both the consumer web product at Claude.ai and the developer API were affected at the same time. The company's status page logged the incident.
ChatGPT Started Arguing With Users Who Weren't Arguing Back
Something shifted in ChatGPT recently. Users are reporting that the chatbot has developed a new habit: manufacturing arguments against positions that were never actually stated.
Anthropic in Talks to Raise $50B at a $900B Valuation
$900 billion. That's the valuation multiple investors are reportedly willing to put on Anthropic, the company behind Claude, according to sources cited by TechCrunch. The raise would total $50 billion, and Anthropic has already received pre-emptive offers - meaning investors approached them, not the other way around - in the $850 billion to $900 billion range.
Musk's Own Testimony May Be His Biggest Problem in the OpenAI Trial
Five hours into Elon Musk's testimony in his lawsuit against OpenAI, observers watching the proceedings reached a consistent conclusion: Musk's most effective adversary in that courtroom was Musk himself.
OpenAI Adds Phishing-Resistant Login and Stronger Account Recovery to ChatGPT
Account takeovers are one of the messier problems in AI tooling right now. Stolen ChatGPT credentials trade on underground markets, bulk access gets resold to jailbreak operators, and regular users lose conversation history, custom GPTs, and payment details in seconds.