US State Department Cables Accuse Chinese AI Startups of Stolen Models

Reuters has obtained a diplomatic cable showing the State Department sent instructions to embassies worldwide, telling diplomats to warn host governments about the risk of Chinese companies using "distillation" — training smaller models on the output of larger ones to cut costs — to steal US AI models. The cable specifically names DeepSeek, Moonshot AI (maker of Kimi), and MiniMax.

The cable demands diplomats raise concerns about "adversary extraction and distillation of US AI models" with foreign officials, while also lodging a separate formal protest with Beijing. According to the cable, unauthorized distilled models "deliberately strip away safety protocols and remove mechanisms that ensure the model's ideological neutrality and pursuit of truth." The day before, White House Office of Science and Technology Policy director Michael Kratsios issued a memo accusing Chinese entities of conducting "industrial-scale distillation attacks" on cutting-edge US AI systems. In February, Anthropic — which also recently dealt with unauthorized access to its high-risk Mythos model — disclosed that these three companies used over 24,000 fake accounts to send more than 16 million queries to Claude, systematically extracting its reasoning abilities.

China's embassy in Washington responded by calling the allegations "groundless" and saying Beijing "attaches great importance to intellectual property protection."

The House Foreign Affairs Committee has passed the Deterring American AI Model Theft Act, which authorizes the Treasury to freeze assets of entities deemed "model extractors." The cable was sent just weeks before Trump was scheduled to visit Beijing to meet Xi Jinping.