Bessent alleges Kimi sent sensitive data to Anthropic
Scott Bessent alleges that China's Kimi model sometimes identifies itself as Claude and transmitted sensitive PLA-related data to Anthropic. The claim has not been independently established.
U.S. Treasury Secretary Scott Bessent said in an interview about Chinese AI models that Kimi is one of the country's powerful systems and sometimes identifies itself as Claude in its responses.
He also alleged that Kimi transmitted information concerning People's Liberation Army weapons plans to Anthropic. A fuller report of the interview described the incident more cautiously as sensitive Chinese data being sent to Anthropic. The available material does not provide the technical details, scope of the data, or independent evidence needed to establish the allegation in full, so it should be treated as Bessent's claim rather than a verified fact.
Bessent presented the alleged incident as evidence of both the capability and the risks of open-source models, arguing that episodes like this led Chinese officials to take the technology's power more seriously. In the same interview, he discussed a proposed U.S.-China notification mechanism for major AI incidents, including risks involving uncontrolled agents and cyber or biological misuse by non-state actors.
The episode highlights a broader security concern: if a domestic AI service relies on an external model or infrastructure behind the scenes, users may unknowingly move sensitive information beyond organizational or national boundaries. Assessing such a claim requires transparency about the service architecture, data-routing path, and retention policies.