NSA, CISA, and FBI published joint advisory AA26-251A naming six Chinese AI firms for industrial-scale distillation of U.S. frontier models
Following·New·first covered 9 Sep 2026·6 sources
NSA, CISA, and FBI name six Chinese AI firms in distillation advisory
The gist
The advisory directs U.S. AI providers to monitor for and respond to suspected distillation using detection patterns that also match legitimate enterprise customers. A White House memorandum has placed covert capability extraction in the national security category, providing policy grounding for further regulatory action.
The full picture
Advisory AA26-251A, issued jointly by NSA, CISA, and FBI, names DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun, and Z.AI as conducting industrial-scale distillation campaigns against U.S. frontier AI models, including Claude, GPT, Gemini, and Grok, since at least late 2024. Specific allegations include Moonshot AI distilling 17 U.S. models to train Kimi-K3, and DeepSeek crafting prompts to extract hidden chain-of-thought reasoning steps. The advisory proposes that U.S. labs serve suspected distillers covertly degraded responses, though the detection indicators also match ordinary enterprise deployments. The White House OSTP's Memorandum NSTM-4, issued April 23, 2026, formally classifies covert, large-scale capability extraction as a national security threat while exempting legitimate distillation. A technical analysis disputes the distillation framing, and a Lawfare analysis urges targeted access controls over new IP rights.
How it developed
Sources
Rohan Paul Twitter
The Neuron
labs.cloudsecurityalliance.org
lawfaremedia.org
1 more source
The daily email
Want this in your inbox?
I send one email each morning with the stories that moved. If you would rather just read here, that works too.
Subscribe free