Anthropic launched Claude Fable 5 on June 9. By June 12, the US government had effectively pulled the plug on global access to it. Three days from debut to shutdown, which might be a record for the fastest rise and fall of a frontier AI model.
The emergency export control directive came after Amazon’s security team flagged a jailbreak vulnerability in the model, prompting federal authorities to order Anthropic to suspend access worldwide.
What happened and why it matters
Claude Fable 5 was not a minor release. It featured a 1M-token context window and was designed for long-horizon agentic tasks, the kind of extended, autonomous AI work that represents the next frontier of capability. Anthropic priced it alongside Claude Mythos 5 at $10 per million input tokens and $50 per million output tokens.
Anthropic had previously invested over 1,000 hours in bug bounty programs without discovering any universal exploits. The vulnerability that triggered the shutdown wasn’t surfaced through those extensive efforts. It came from Amazon’s security team, which flagged a specific jailbreak concern serious enough to catch federal attention.
