US Shut Down Fable 5 and Mythos 5 at Anthropic - AI Switched Off for the Whole World
Commerce Department: ban for foreigners. Anthropic killed the models globally. IPO at $965B at risk. Jailbreak or politics?
Anthropic launched Fable 5 on June 9. On June 12 the US government said: shut it down. By June 13 the models were gone for everyone - not just overseas.
The Commerce Department issued an export-control directive: block access to Fable 5 and Mythos 5 for any foreign national - in the US and abroad. Anthropic got the letter at 5:21 PM Eastern on Friday. To stay on the right side of the wording, the company turned off both models globally for all customers.
The official reason is national security. The administration points to a jailbreak: a way to bypass restrictions and make the AI hunt for code vulnerabilities. Amazon and five other companies tested Fable 5 and found a hole that led to Mythos 5 - the “cybersecurity” version of the model.
Anthropic pushed back hard:
- the jailbreak is narrow, not universal
- the same vulnerability exists in GPT-5.5 from OpenAI “without any hacking”
- if you apply that standard across the whole industry - “basically every new frontier release stops”
- evidence from the government so far is only verbal
David Sacks (White House science adviser) wrote that he asked CEO Dario Amodei to patch or pull Fable 5 - he refused. Amodei flew to Washington for meetings.
The scandal has wider context:
- the Pentagon earlier put Anthropic on a blacklist - the company would not give the military unlimited AI access for weapons and surveillance
- Anthropic is suing the government
- in early June it filed for an IPO at a ~$965B valuation - now uncertainty
Europe is nervous: the European Commission talks about tech sovereignty and says measures must not be discriminatory toward allies. For the first time an AI model was effectively treated like an export-controlled good.
Regular Claude models still work. But the top tier is dead. Millions of users who had just gotten used to Fable are back without the “smartest” mode.
The question is not just about security. This is a fight over who controls AI - labs, the Pentagon, or a Friday evening decree from the Commerce Department.