The White House Just Shut Down Anthropic’s Most Powerful AI Models Over a ‘Jailbreak’ That Wasn’t What It Seemed
Export controls imposed at 5:23pm Friday, Anthropic scrambles to Washington as models remain paused.
On June 18, 2026, the White House imposed export controls at 5:23pm Friday, shutting down Anthropic's Claude Fable 5 and Claude Mythos 5. The action, triggered by a jailbreak report from Amazon, caught Anthropic off guard. CEO Dario Amodei flew to Washington to meet with the Trump administration on Monday, hoping for a quick resolution. However, the administration insists the models stay offline until the jailbreak is 'fixed'—a demand Anthropic considers technically impossible.
The jailbreak in question involved instructing Fable to 'fix this code'—a task it performed willingly, identifying security vulnerabilities that could then be exploited. The same weaknesses were already easily found by Opus 4.8 and GPT-5.5. Anthropic argues that blocking such defensive coding would require crippling the model's ability to write secure code entirely. The pause continues into its seventh day, with betting markets giving under even odds for resolution by July 1. Meanwhile, other AI news includes MidJourney Medical's full-body scanner and DeepSeek raising $7.5B.
- White House export controls shut down Claude Fable 5 and Mythos 5 at 5:23pm Friday after Amazon reported a jailbreak.
- The jailbreak involved the model fixing code vulnerabilities—a capability Anthropic says cannot be removed without crippling coding ability.
- Pause is now day seven; Anthropic met with Trump administration but faces impossible demands to 'fix' the issue.
Why It Matters
Government intervention halts frontier AI deployment over security concerns, setting a precedent for model regulation.