Claude Fable 5 is running again. Anthropic's most capable public model came back on July 1 — two and a half weeks after it was abruptly switched off on orders from the US government.
The reason was serious. In mid-June, researchers at Amazon managed to get past Fable 5's built-in guardrails and pushed it to find real software vulnerabilities and write code to exploit them. In response, the US Commerce Department slapped sudden export controls on Fable 5 and its locked-down "older sibling" Mythos 5, citing national security.
Formally, Anthropic was ordered to cut off access for "any foreign national — inside or outside the United States, including its own foreign-national employees." The company had no way to verify users' nationality in real time, so it had to shut both models down worldwide to stay within the rules.
To bring the model back, Anthropic spent two weeks building a new safety classifier that blocks the exact exploit technique Amazon found — with 99% accuracy. The Commerce Department reviewed the models alongside the company throughout. In return, Anthropic agreed to hunt for vulnerabilities itself, coordinate future launches, and report any attempts at misuse. The controls were lifted on June 30, and access returned on July 1.
The episode shows how tightly frontier AI is now tied to the government. One successful jailbreak, and a powerful model is switched off worldwide for two weeks. For Anthropic it is both a lesson and the price of leadership: the stronger the model, the closer the regulators.
