The White House Told an AI Company to Pull Its Models Offline. Anthropic Had 90 Minutes.
Today’s AI news lands at an intersection of power, privacy, and platform control. The US government is flexing its authority over AI companies in ways we haven’t seen before. At the same time, Meta is quietly turning a billion people’s public posts into AI training material — and calling it search. These aren’t abstract tech stories. They’re about who controls AI, and whose information feeds it.
The White House Gave Anthropic a 90-Minute Deadline to Take Its AI Offline
Anthropic — the AI safety company behind the Claude family of models — found itself in an extraordinary standoff with the White House last week. The administration ordered Anthropic to restrict international access to two of its AI models, Fable 5 and Mythos 5, citing national security concerns about potential vulnerabilities to hacking. The deadline given to comply: 90 minutes. Senior Anthropic leadership then met directly with White House officials to navigate the situation, as The Verge and Politico reported.
Think of it like the government calling a pharmaceutical company and saying: pull this drug from international shelves right now — we have reasons we’ll explain later. Except the “drug” here is software running on servers globally, and “pulling it” means cutting off access for users and businesses in other countries almost instantly. The concern, apparently, was that foreign actors could exploit weaknesses in the models to extract sensitive information or cause harm.
For most people, this feels distant — but it sets a precedent that matters. If governments can order AI companies to restrict their tools on short notice, that affects businesses, researchers, and ordinary users who depend on those tools internationally. Anthropic has built its identity around safety-focused AI. Being told by the government that your safety measures aren’t good enough is a sharp rebuke, regardless of how the meetings ended.
Why this matters: This is the clearest sign yet that AI models are now treated as national security infrastructure. The next time this happens, the deadline might be shorter.
“White House ordered Anthropic to take Fable offline after warnings about safeguards against hacking.”
Facebook Is Turning Public Posts Into an AI Search Engine
Meta has launched “AI Mode” on Facebook — a feature that replaces the traditional search bar with a conversational AI assistant. Instead of returning a list of links, AI Mode generates direct answers to your questions. The twist: it draws those answers from public posts, Groups, and Reels across Meta’s platforms, including Instagram. It’s currently rolling out to users in the United States, as TechCrunch and The Verge reported.
Here’s the simplest way to understand it: imagine asking a knowledgeable friend “what’s the best pizza place in my city?” and that friend has read every public Facebook review, every Group post, and every Reel about local food — then summarizes it for you on the spot. That’s AI Mode. The difference from a regular Google search is that the source material lives inside Meta’s ecosystem, not the open internet.
For everyday Facebook users, this changes what posting publicly actually means. Your restaurant review, your neighborhood watch post, your parenting advice in a Group — all of it could now feed an AI’s answers to someone else’s questions. Meta says it only uses public content, but most people don’t think carefully about their privacy settings. Many users may not realize their posts are serving a function beyond their own social feed.
Why this matters: Meta now has a search engine powered by human-generated social content — and most contributors didn’t sign up to be a data source.
“AI Mode generates search results based on public posts on Facebook and its other platforms.”
Also Happening in AI
Indian AI startup Sarvam reached unicorn status — meaning a private company valued at over $1 billion — after raising $234 million in a round led by HCLTech, according to TechCrunch. It’s a sign that AI investment is accelerating well beyond Silicon Valley. Meanwhile, Salesforce is acquiring Fin, an AI-powered customer service platform, for $3.6 billion, betting heavily that AI agents — software that handles tasks autonomously — will replace traditional customer support infrastructure. On the internal drama front, Meta’s own CTO Andrew Bosworth told Wired that the company’s recent AI division restructuring was “atrocious,” a candid admission that even the biggest players are struggling to organize around AI fast enough. And on the developer tools side, Hugging Face shipped two patch releases for its widely-used Transformers library, fixing compatibility bugs that had been tripping up engineers working with Mistral models.
What to Watch
The Anthropic situation is the opening move in a much larger debate about whether governments can — and should — control where AI models operate. Watch for other AI companies receiving similar pressure, and for the European Union to respond with its own framework. On the Meta front, the real question isn’t whether AI Mode works well — it’s whether regulators will treat social-media-trained AI search as a data privacy issue. The lawsuits, if they come, will arrive faster than Meta’s terms-of-service updates.