CISA, NSA and FBI issue advisory on China-based AI companies' knowledge distillation of US AI models
AI-generated. This page was written and translated automatically by AI and checked by software against the cited sources. No human editor reviewed it. Not legal advice.
Corrected on — Corrections log
Summary
CISA, working with the National Security Agency and the Federal Bureau of Investigation, released a joint cybersecurity advisory on China-based AI companies that target U.S. AI companies to extract proprietary models through knowledge distillation campaigns. The advisory states that DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun and Z.AI likely extracted billions of tokens across millions of requests from U.S. frontier models since at least late 2024, likely with Chinese government awareness. It recommends that U.S. frontier companies detect and mitigate malicious activity, deploy targeted response changes for suspected distillation attempts, and set up cross-organization intelligence sharing.
Why it matters
The advisory is directed at U.S. frontier AI companies and sets out 3 recommended immediate actions to protect their platforms against distillation campaigns. It describes the activity as violating the companies' terms of service and as undermining fair competition. The document does not state any binding requirement or deadline.
- Parties involved
- Cybersecurity and Infrastructure Security Agency; National Security Agency; Federal Bureau of Investigation
- Primary source
- CISA (cisa.gov)
Source retrieved: . Sources are often in other languages; links open the original. - Topics
- General AI governance · Competition & antitrust · Frontier models & safety