Anthropic's AI pulled by government, Meta's engineers in revolt, Mistral doubles its valuation
The government just killed Anthropic's most powerful models, Meta's 6,500-person AI unit is imploding, and Mistral is quietly becoming a $23B juggernaut.
Subscribe free →The Government Yanked Anthropic's Best Models — And Anthropic Is Furious
A government order forced Anthropic to completely cut off access to its Fable 5 and Mythos 5 models for all customers — including its own employees — citing national security concerns over a discovered jailbreak vulnerability. Anthropic pushed back hard in a public statement, arguing that "a narrow potential jailbreak should not be cause for recalling a commercial model deployed to hundreds of millions of people." The irony is brutal: Anthropic built its entire brand around safety, and that credibility may have made regulators more willing to pull the trigger.
Why it matters: This is the first time a government has ordered a major AI lab to shut down commercial model access, and it sets a precedent that could reshape how frontier AI is deployed and regulated going forward.
Meta's 6,500-Person AI Division Is Eating Itself Alive
Engineers inside Meta's consolidated AI unit are describing it as a "soul-crushing gulag," with morale cratering and the organization reportedly on the verge of open revolt. The unit, which was formed just months ago by folding together multiple AI teams, has grown to 6,500 people — and apparently no one knows what they're supposed to be building or why. Centralization was supposed to make Meta's AI push more focused; instead it seems to have created a bureaucratic nightmare at scale.
Why it matters: When your AI talent pipeline starts describing their workplace like a Soviet labor camp, you have a retention and execution problem — and Meta can't afford that with OpenAI, Google, and Anthropic all poaching aggressively.
Mistral Is Quietly Building a $23 Billion Empire
Europe's AI darling Mistral is rumored to be raising €3 billion at a €20 billion (~$23B) valuation — nearly double the €11.7B valuation it commanded at its Series C. No deal has been confirmed yet, but if it closes at these numbers, Mistral will have pulled off one of the fastest valuation doublings in recent AI history. The company has been steadily expanding its enterprise footprint while staying lean compared to its American rivals.
Why it matters: A $23B Mistral signals that European AI is no longer just a geopolitical talking point — there's real capital conviction behind an open-weight competitor to GPT and Claude.
Google Sues Chinese AI Scam Ring That Blasted 2.5 Million Texts in Two Weeks
Google has filed a lawsuit against a Chinese cybercrime operation called "Outsider Enterprise," which allegedly used AI tools to scam hundreds of thousands of victims via SMS — sending 2.5 million fraudulent text messages in just two weeks. The suit is one of the most aggressive legal actions a tech company has taken against AI-powered fraud operations. Google is using the lawsuit to establish legal precedent and gather discovery on how exactly the AI tools were weaponized.
Why it matters: AI-supercharged fraud is scaling faster than law enforcement can respond, and this lawsuit is a signal that big tech is going to start taking legal offense rather than just playing defense.
Siri AI Is Finally Worth Turning Back On
The Verge spent 24 hours with the new Siri AI shipping in macOS 27 Golden Gate and came away cautiously impressed — notable for a reviewer who had Siri completely disabled and found Apple Intelligence "fruitless." The new version appears to be a meaningful step up, though it's still in early developer preview and the reviewer was careful not to declare victory. Apple has a long runway to close before Siri is genuinely competitive with Claude or Gemini, but at least it's no longer embarrassing.
Why it matters: If Apple's 1B+ active device install base actually starts engaging with a competent Siri AI, the downstream effects on AI assistant adoption — and on Apple's platform lock-in — are enormous.
Hollywood's AI Future Isn't Plug-and-Play Prompting — It's Custom-Trained Models
A new piece out of Tribeca 2026 digs into how serious filmmakers are actually using AI: not by dumping prompts into off-the-shelf video generators, but by training custom builds of models like Google's Veo and Imagen on their own concept art and visual language. The project spotlighted, *Dear Upstairs Neighbors*, represents a different vision of AI filmmaking than the "press generate and hope" workflow being hyped everywhere. Craft still matters — AI just changes what the craft looks like.
Why it matters: Custom-trained models for specific creative visions, not generic generators, is likely the actual path to AI content that audiences will pay to watch.
Quick Hits
- →Apple's new iOS 27 AI photo editing tools — Reframe, Extend, Clean Up — are now in hands-on testing and reviewers say they mostly work, putting the iPhone closer to Pixel territory for the first time. The Verge
- →A critical 0-day in Oracle's PeopleSoft is actively being exploited to steal gigabytes of data from hundreds of organizations — patch immediately if this is in your stack. Ars Technica
- →A Verge writer vibe-coded a backyard plant management app using Gemini with a single long prompt — bugs and all — and it mostly worked, which is either inspiring or terrifying depending on your job. The Verge
- →The FBI built a full replica small town inside an Alabama building to simulate real-world cyberattacks on infrastructure like power grids and water systems. TechCrunch
AI TLDR