Inside the fight over Claude Mythos 5
While America was celebrating a World Cup win and a Knicks championship, Anthropic was in a full-blown crisis with its own government — forced to pull Fable 5 and Mythos 5, its most powerful models to date, after a surprise export control directive landed on a Friday afternoon. The timing is almost too on-the-nose: the country was partying, and one of its leading AI companies was quietly being kneecapped. The models had been live for less than a week.
Trump’s Anthropic shutdown just made the case for non-American AI
This is the story that will get quoted in board meetings in Paris, Seoul, and Riyadh for the next year. If a US company can be ordered to cut off its own foreign employees overnight, every government on earth just got a very clear signal about the risks of depending on American frontier AI. The irony is exquisite: Washington’s attempt to control AI exports may have just done more to accelerate sovereign AI movements abroad than any amount of EU policy ever could.
Anthropic Is Still at Odds With the White House Over Claude Fable 5
Anthropic’s leadership flew to DC Monday for face-to-face talks with White House officials — and left without a resolution. Both sides remain split on what risk Fable 5 actually presents, which tells you one of two things: either Anthropic genuinely built something the government finds alarming, or this whole episode is as politically motivated as it looks. Given the companion reporting that this was never actually about a jailbreak, I know which way I’m leaning.
Cybersecurity vets protest ‘dangerous’ US government ban on Anthropic’s most powerful models
Dozens of cybersecurity professionals are now pushing back on the White House, arguing that pulling Fable 5 and Mythos 5 hurts defenders far more than attackers. This is the classic dual-use dilemma in extremely sharp relief — the same model that could help someone write malware can also help patch it, and the people best positioned to know that are the ones signing this letter. Good luck getting the administration to read it carefully.
DOJ Lawyers Argue xAI Is ‘Vital’ for National Security in NAACP Lawsuit
The Justice Department is now arguing in court that Elon Musk’s xAI is so essential to US military operations — including, yes, the Iran War — that a lawsuit over its polluting gas turbines in Memphis should be dismissed on national security grounds. Let’s be clear about what’s happening here: the federal government is using “military necessity” to shield a private AI company’s environmental violations from a civil rights lawsuit. The merger of government power and billionaire tech ventures is proceeding right on schedule.
Meta Tapped a Pentagon Supplier to Prototype Face Recognition for Its Glasses
Meta worked with Rank One Computing — a firm whose board includes a former CIA deputy director and a former FBI science chief — to prototype face recognition for its Ray-Ban smart glasses. This wasn’t a product launch; it was internal development, which actually makes it more interesting. The gap between “this is just a fashion accessory” and “this can identify strangers on the street” is apparently a lot smaller than Meta’s marketing department would prefer you to know.
It Is Trivially Easy to Use Reddit to Manipulate AI Search, Research Suggests
New research finds that just 13 words of planted text on Reddit, Wikipedia, or Quora can reliably steer AI agents into outputting spam and scam content. Thirteen words. Half a tweet. The entire industry has spent billions building systems that can be poisoned at the source for essentially nothing, and the source is the same user-generated content those systems were trained to trust. This is not a theoretical vulnerability — it’s the current state of play.
Big Tech’s desperate last push at AI regulation
The tech industry’s lobbying apparatus is making its final push for federal AI preemption — a single national framework that would override the growing patchwork of state laws. The pitch is framed as simplification and innovation, but the subtext is obvious: one federal law is much easier to shape than fifty state ones. Whether Congress bites before the midterm window closes is another question entirely, but the Anthropic saga this week will give both sides new ammunition.
Bottom Line
The US government just reminded the entire world that “American AI dominance” is a policy choice that can be reversed on a Friday afternoon — and the world was paying attention.