U.S. halts Anthropic AI model releases over security flaw

What happened

Amazon researchers found a jailbreak in Anthropic's Fable 5 model that could bypass safety protections and expose cybersecurity risks. Amazon CEO Andy Jassy mentioned the vulnerability to Treasury Secretary Scott Bessent during a June 11 call about an unrelated topic. Within four days, Commerce Secretary Howard Lutnick issued export controls requiring government approval before Anthropic could release Fable 5 and Mythos 5 to foreign nationals or users abroad, with a 90-minute deadline. Anthropic took both models offline by 10 p.m. that Friday.

Why it matters

This is the first time the U.S. government has stepped in to explicitly limit the release of a frontier AI model, setting what may be a new precedent for how Washington regulates AI companies. For Anthropic—which recently closed a $65 billion（約10兆円） funding round at a valuation of $965 billion（約150兆円） and has filed to go public—the export controls represent an existential threat at a time when rivals OpenAI, SpaceX, and Google are competing aggressively. Cybersecurity experts warn that government licensing of AI models could make U.S. labs less competitive and allow geopolitical rivals like China to gain ground.

What to watch

The two sides remained at an impasse as of Tuesday, with export controls still in place and next steps unclear. Anthropic maintains that the jailbreak was relatively simple and could be achieved using other available models, but the administration argues the bypass unlocks dangerous cyber capabilities. An Anthropic spokesperson said both parties are working quickly to resolve the issue, but no resolution has been announced.