The NSA, CISA and FBI have issued a joint cybersecurity advisory.
American authorities are accusing Chinese companies of “distillation activities at an industrial scale.” The National Security Agency (NSA), Cybersecurity and Infrastructure Security Agency (CISA) and Federal Bureau of Investigation (FBI) have issued a joint cybersecurity advisory, warning organizations that Chinese AI companies are extracting the “proprietary functionalities and capabilities” of frontier American models.
Specifically, the agencies are accusing DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun and Z.AI of extracting “billions of tokens across millions of exchanges/requests” from American models, including Anthropic‘s Claude, OpenAI’s GPT, Google’s Gemini and xAI’s Grok, since 2024. Distillation is the term for a method in AI development in which the output of a powerful model is used to train newer ones.
The warning states that DeepSeek used data and capabilities from various Claude, Gemini, GPT and Grok models to train its own, including the R1. That’s the powerful, open-source reasoning model the company released in early 2025. Moonshot AI, the agencies said, extracted “significant” data from Claude’s Fable to train Kimi K3, which is considered one of the most advanced models developed by a Chinese company. Anthropic created Fable to bring some of the capabilities of Mythos, its state-of-the-art cybersecurity model exclusively available to Project Glasswing participants, to the public. Similarly, Moonshot’s Kimi K2 used GPT-4o data, the agencies add.
The warning lists the American frontier models the Chinese companies allegedly used to train their own, along with the mitigations US firms can implement to combat “malicious distillation campaigns.”
American companies have long alleged that Chinese firms are trying to copy their technologies. Shortly after DeepSeek became the top free iPhone app early last year, OpenAI said that, along with Microsoft, it had been banning accounts it suspected of “distilling” its technologies. DeepSeek was one of the companies it investigated. Earlier this year, Anthropic also accused DeepSeek, Moonshot and MiniMax of wide-scale distillation campaigns. But it’s not just Chinese companies: In OpenAI’s announcement about pulling its models from Cursor, it noted that Elon Musk admitted to using OpenAI’s output to train xAI models during cross-examination in the lawsuit he filed against the company.
