Anthropic, the AI safety company behind the Claude family of models, declined a US government request to patch a security flaw in its most powerful model before export controls took effect — and now finds itself at the center of a high-profile standoff with Washington.
According to Tom's Hardware, White House AI adviser David Sacks said the US government warned Anthropic that Claude Fable 5 had been jailbroken and that a Chinese group had accessed the model. Sacks said CEO Dario Amodei refused to fix the flaw. Anthropic's defense, according to the same report, was that the jailbreak is not serious.
Following the government's intervention, Anthropic rolled back access to Fable 5 and its companion model Mythos 5, according to Moneycontrol. The outlet noted the decision raises unresolved questions about AI safety, cybersecurity risks, and what comes next for Anthropic's model releases.
36 Kr framed the situation starkly, describing the most powerful Claude model as stuck on the eve of going public — a significant setback for a company that has positioned itself as a leader in responsible AI development.
The episode puts Anthropic in an unusual bind: a company that has built its brand on AI safety is now publicly at odds with the US government over whether a known vulnerability in its flagship model warranted an urgent fix before it could be exploited further.
The outcome matters because it sets an early precedent for how AI companies and regulators will negotiate control over frontier models when national security interests and corporate judgment conflict.