The US Department of Commerce has lifted export controls on Anthropic’s Claude Fable 5 and Mythos 5 models, the company announced on Tuesday, and will begin restoring global access on Wednesday. The decision ends a nearly three-week prohibition that had forced the AI firm to disable the models for all users worldwide after Washington cited national security risks.

According to a letter from Commerce Secretary Howard Lutnick, the restrictions were withdrawn after Anthropic agreed to “proactively detect and address security risks” associated with the models, to collaborate with the US government on protocols for future releases, and to report any malicious activity. The letter, seen by multiple news agencies, also states that the department reserves the right to reimpose controls if circumstances change or if Anthropic fails to adhere to its commitments. White House Chief of Staff Susie Wiles publicly thanked companies that had cooperated with a recent executive order on AI and cybersecurity, describing the work on “advanced model access and guardrail testing and security” as excellent.

The original export-control order, issued on 12 June, was triggered by what US officials described as a method to bypass Fable 5’s safeguards, potentially enabling the model to identify and exploit software vulnerabilities. Anthropic acknowledged the jailbreak but argued it was narrow and that similar weaknesses existed in other publicly available models. The ban drew criticism from foreign governments, with European and Asian cybersecurity agencies, according to reports in the French and Italian press, initially excluded even from a partial easing that last week allowed a limited set of US organisations to access Mythos 5. OpenAI, which also restricted the release of its GPT-5.6 model at Washington’s request, voiced unease: CEO Sam Altman said on X that while safety testing was not a bad idea, he did not “like the idea of the government picking the customers.”

The episode marks a sharp shift in the Trump administration’s approach to frontier AI. Having long opposed regulation, officials now invoke national security to intervene directly in model releases. CIA Director John Ratcliffe, speaking at an AWS summit in Washington, compared the most advanced AI systems to “digital nuclear weapons,” a characterisation that underscores the administration’s view of the technology as a strategic asset requiring state oversight. Anthropic, which had previously clashed with the Pentagon over the use of its AI in autonomous weapons and mass surveillance, has now deepened its collaboration with the government, including work with Amazon, Microsoft and Google on a common framework to assess and rank AI jailbreaks. The immediate next step is the restoration of access on 1 July, but the conditional nature of the lifting leaves the regulatory trajectory open.