Anthropic clashes with Trump admin over Claude Fable 5 export controls
Jailbreak fears lead to emergency talks and model shutdown
Anthropic's emergency talks with the Trump administration on Monday concluded without resolving export controls imposed on Claude Fable 5, a version of the company's Mythos model. The White House, spurred by a call from Amazon CEO Andy Jassy to Treasury Secretary Scott Bessent, believes the model's guardrails can be jailbroken to unlock advanced cybersecurity capabilities. The NSA concurred, leading to restrictions. Anthropic argues the concerns are overblown and has shut down consumer access to Fable 5. The Commerce Department, led by Secretary Howard Lutnick, indicated willingness to restore access if jailbreak issues are fully resolved.
A group of cybersecurity researchers sent an open letter supporting Anthropic, stating that Mythos-class models are not uniquely dangerous and that the controls undermine defenders and risk America's AI leadership. The dispute occurs amid broader tensions between Anthropic and the Pentagon over military use of its AI. Anthropic's cofounder Tom Brown and head of external affairs Sarah Heck are leading discussions, with researchers Logan Graham and Nicholas Carlini in Washington. The outcome remains uncertain, with both sides claiming a desire for resolution.
- Trump administration imposed export controls on Claude Fable 5 after Amazon CEO Andy Jassy alerted Treasury Secretary Scott Bessent about jailbreak vulnerabilities.
- NSA confirmed guardrails could be bypassed, but Anthropic and cybersecurity researchers argue the risks are overblown and the action harms AI defenders.
- Talks at Commerce Department involving Secretary Howard Lutnick failed to lift controls; Anthropic has cut off consumer access to Fable 5 pending resolution.
Why It Matters
This spat over model safety threatens US AI leadership and creates market uncertainty for enterprise AI adoption.