The US government has forced Anthropic to withdraw its two newest AI models, Fable 5 and Mythos 5, citing national security concerns. According to TechCrunch, the move came just as last week was ending, after Amazon researchers allegedly found a way to bypass Fable 5's guardrails.
Guardrails are the safety controls AI companies build to stop a model from producing dangerous output — and a "jailbreak" is any trick that gets around them. The reported Amazon finding suggested those protections could be defeated, which appears to have triggered the government's intervention.
The demand raises a thorny technical question. As Gadget Review frames it, the White House wants Anthropic to block all jailbreaks — something the outlet suggests may simply be impossible. Security experts widely consider fully jailbreak-proof models an unsolved problem, so a standard of zero successful bypasses could be one no model can currently meet.
The pushback has already begun. According to TechCrunch, cybersecurity researchers have signed an open letter responding to the ban, though the available reporting cuts off before detailing their full argument.
There is also a counterintuitive twist. TechCrunch reports that the ban may be accidentally helping Anthropic's brand, noting in a separate piece that despite the forced withdrawal of the Fable 5 release, "the numbers don't seem to care" — a sign the company's standing or usage has not visibly suffered.
Why it matters: This is a rare case of the US government directly ordering a leading AI company to pull products from the market, and the demand to eliminate every jailbreak could set a safety bar that no current AI model is able to clear.