The US government forced Anthropic to pull its most powerful AI models, a move that looks partly about safety and partly about politics — bad for Anthropic, and a sign of tighter AI control ahead.

• The government issued an export order blocking foreign nationals from using Anthropic's new Fable 5 and Mythos 5 models, which forced Anthropic to shut them down for everyone.
• The crackdown reportedly started after Amazon's CEO told officials his researchers tricked Fable 5 into giving cyberattack info it was supposed to block.
• Anthropic says it was unfairly singled out, since no one found a full "universal jailbreak" and rival models carry similar risks.
• Critics see politics at play: Anthropic's boss Dario Amodei has not cozied up to Trump like Elon Musk or Sam Altman, while a broader safety rule for all AI companies was reportedly killed by David Sacks.
• The deeper worry is "recursive self-improvement" — AI that trains better AI on its own — which Anthropic wants paused, though that also conveniently protects its lead.

Outlook: Officials call the issue fixable and say the ball is in Anthropic's court, so access may return once the flagged flaws are patched.