The US National Security Agency (NSA), the Cybersecurity and Infrastructure Security Agency (CISA), and the Federal Bureau of Investigation (FBI) have announced that six Chinese AI companies are conducting large-scale distillation to extract capabilities from cutting-edge US AI models. The accused companies include DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun, and Z.AI.

According to the authorities' announcement, these companies have been executing millions of queries since at least late 2024 to extract advanced inference capabilities and specific features from models such as Claude, GPT, Gemini, and Grok. It was pointed out that these activities have allowed the Chinese firms to significantly reduce computational costs and research time in model development.

Specific methods cited include bypassing geographical restrictions using proxies, the mass use of APIs through fake accounts, and the extraction of thinking processes (Chain-of-Thought) via prompt injection. The authorities recommend that US AI companies monitor suspicious accounts and implement countermeasures, such as intentionally degrading response accuracy when distillation is suspected.

Source: