Bessent says Chinese AI labs steal from US models, names Kimi

US Treasury Secretary Scott Bessent said on The Axios Show that China's AI models "do industrial distillation, which is a nice word for stealing from the U.S. models." Axios posted the clip on Saturday, with the interview conducted by Mike Allen. The remarks are Bessent's own characterisation, and the clip carries no evidence for the claim.

Bessent singled out one system by name. "One of their very powerful models is Kimi," he said. "Kimi thinks she is Claude. Kimi will tell you she is Claude." He did not say how the Treasury reached that view. Axios's post does not say whether he relied on a government assessment or on users' reports of the model's behaviour.

https://x.com/axios/status/2106416485992415256

In the same conversation, according to a post by the AI-policy commentator Andrew Curran quoting the exchange, Axios asked whether the administration might become more interventionist on AI after the midterms. Bessent replied that "we want safe acceleration" and that it depended on what was meant by activist and interventionist.

Bloomberg reported that Bessent also criticised prominent AI figures' warnings about existential risk as alarmist and unhelpful, and called on the industry to police itself and produce solutions. Curran's post gives his wording as "alarmism without solutions by some of the AI community, that's not leadership." Neither Moonshot AI nor Anthropic had commented in the sources reviewed.