Treasury Secretary Scott Bessent accused China of stealing AI models from the U.S. by distillation —"a nice word for stealing"—and revealed that Kimi "thinks he's Claude" and that he returned PLA weapons plans to Anthropic. He told Axios on Oct. 3 and opened the door to a U.S.-China emergency channel. for AI incidents.
The context that Axios documents: China "woke up" in 60 days to the power of its open models; distillation trains a smaller model with outputs of a larger one, in fraction of time and cost. Fireside Alpha amplified the interview with Mike Allen: communication channel yes, and responsibility in laboratories (Huttenlocher dixit).
You may also likeAnthropic apunta a 1,8 billones de dólares en su debut en NasdaqThe Kimi case has a history: in July Moonshot's K3 equaled Claude and ChatGPT in coding (Arena), and in February Anthropic had already accused DeepSeek, Moonshot and MiniMax of illicit distillation. Beijing denies everything ("unfounded"); the ironic twist is Cursor, built on Chinese K2.5, in the process of being purchased by SpaceX.
The phrase that itches the most: a Chinese model returning weapons plans to the American laboratory from which it would have drunk. If it is literal, distillation is no longer just commercial theft: it is reverse military transfer.
The punchline is institutional: Bessent proposes security notifications between powers and attributing responsibility. Washington and Beijing talking about rules for AI that both accelerate: tactical détente or real protocol, will be seen in the first crisis.
For the industry, the message is twofold: take care of the pesos like gold and assume that all public output trains the rival. The era of the open and innocent model ended this week.
You may also likeMicrosoft services almost restored after 24 hours of outageImage credit: Great Wall, Velatrix via Wikimedia Commons, under CC0 license.




