The US Commerce Department last week banned exports of two Anthropic AI models — Claude Fable 5 and Mythos 5 — cutting off access to users outside the country and foreign nationals inside it. The move followed President Trump telling Axios he viewed the company as a national security threat, though relations have since warmed after CEO Dario Amodei responded quickly.
Why the models were blocked
The export ban came on the heels of a vulnerability alert from Amazon, a competitor and part-owner of Anthropic. Amazon flagged the issue to the administration, though details about the exact flaw remain undisclosed. Trump confirmed the alert in his Axios interview, saying it prompted the initial national security concern. Anthropic's CEO later said that some early testers of the Mythos model described it as a 'super weapon' and suggested it should require licensing.
Joint risk framework takes shape
The White House and Anthropic are now drafting a joint risk framework to assess how severe AI security flaws are and when the government should step in. Negotiators aim to define three things: how far safeguards were bypassed, what capabilities were exposed, and the real-world consequences of any breach. The framework could eventually serve as a template for how other governments and AI developers handle similar risks across the industry.
Trump: AI race with China outweighs political clashes
Trump made clear that beating China in artificial intelligence is a higher priority than any political friction with Anthropic or its peers. He said the US is winning by a lot, based on his conversation with President Xi, and that while he wouldn't necessarily use the Defense Production Act to control national AI, he thinks the industry has acted responsibly so far.
Technical standards still up in the air
The relationship between the company and the White House is on the mend, but the technical work on AI safety standards remains uncertain. Negotiators are still hashing out what counts as a serious breach and how to measure real-world impact. Until those details are nailed down, the practical effect of any safety pact will be hard to gauge.
