NSA, CISA and FBI Warn of "Industrial-Scale" Distillation by Chinese AI Firms
A joint US advisory names DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun and Z.AI, alleging they extracted billions of tokens from American frontier models since 2024.
The National Security Agency (NSA), the Cybersecurity and Infrastructure Security Agency (CISA) and the FBI have published a joint cybersecurity advisory alleging that Chinese AI companies are extracting the "proprietary functionalities and capabilities" of American frontier models, in what the agencies call "distillation activities at an industrial scale."
The advisory names DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun and Z.AI. According to the agencies, these companies have pulled "billions of tokens across millions of exchanges/requests" from American systems since 2024, including Anthropic's Claude, OpenAI's GPT, Google's Gemini and xAI's Grok. Distillation refers to a training method in which the output of a powerful model is used to train newer models.
Specific claims
The document says DeepSeek used data and capabilities from various Claude, Gemini, GPT and Grok models to train its own systems, including R1, the open-source reasoning model released in early 2025. Moonshot AI, the agencies say, extracted "significant" data from Claude's Fable to train Kimi K3. Anthropic built Fable to bring some capabilities of Mythos, its cybersecurity model available only to Project Glasswing participants, to a wider audience. The agencies add that Moonshot's Kimi K2 used GPT-4o data.
The advisory lists the American frontier models allegedly used as training sources and sets out mitigations US companies can apply against "malicious distillation campaigns."
Earlier accusations
Similar claims have circulated before. Soon after DeepSeek became the top free iPhone app early last year, OpenAI said it and Microsoft had banned accounts suspected of distilling its technology, with DeepSeek among the companies investigated. Anthropic also accused DeepSeek, Moonshot and MiniMax of large-scale distillation earlier this year.
The issue is not limited to Chinese firms. In its announcement about pulling its models from Cursor, OpenAI noted that Elon Musk admitted under cross-examination, in the lawsuit he filed against the company, to using OpenAI output to train xAI models.
Full report on Engadget.
