NSA, FBI and CISA Jointly Accuse Six Chinese AI Firms of Industrial-Scale Model Distillation

On September 8, 2026, the NSA, FBI and CISA issued joint advisory AA26-251A accusing Chinese AI companies of industrial-scale knowledge distillation of US frontier models: systematically extracting restricted proprietary capabilities and training their own systems on the outputs, thereby closing the capability gap without bearing frontier-level compute, power and R&D costs. The agencies warned such activity could strengthen China's military and cyber capabilities. This is the first time the three agencies have formally framed model distillation as a national security issue.

Confirmed

• The advisory, numbered AA26-251A and titled on China's AI companies' industrial-scale distillation campaign against US AI companies, was published on the CISA website on September 8, 2026; @rohanpaulai and @castrotech both linked the original document.
• The executive summary states that distillation is core rather than supplementary to the named companies' AI development strategy; @chrisrohlf relayed the report's characterization as systematic extraction of proprietary capabilities threatening US technological leadership.
• Per @rohanpaulai and @chrisrohlf, the advisory names DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun and Z.AI as conducting the campaign since late 2024, using gray-market channels among other means; @DynamicWebPaige confirms a six-company list (one anonymized) and cites "billions of tokens extracted via millions of requests."
• Per a Kalshi bulletin relayed by @ChrisUniverse, the FBI alleges copying of Claude, GPT, Gemini and Grok; @ChrisUniverse quipped this is old news.
• @VraserX relayed a version naming only DeepSeek, Moonshot and Alibaba, and philosophically questioned where the line of permissible learning from model outputs lies.

Unconfirmed

• A Polymarket bulletin stresses the details await further verification; the company lists vary across posts (six vs three names, with the sixth given as either Z.AI or an anonymized firm) and have not been checked line-by-line against the advisory text.

Why it matters

• This is the first joint advisory elevating model distillation from a commercial compliance dispute to an official national security accusation tied to military and cyber risk, with potential implications for export controls and API usage policy.
• @VraserX's question about the boundary of learning from outputs signals that any enforceable rule would first require a much clearer technical definition of permissible distillation.

2026-09-09 ~ 2026-09-11 · 10 related posts

• NSA, FBI, CISA advisory accuses Chinese AI firms of industrial-scale distillation of US frontier models — Cklly2004 · 2026-09-09
• U.S. accuses Chinese AI firms of 'malicious' AI distillation at industrial scale — Polymarket · 2026-09-09
• [source] NSA, CISA, FBI accuse 6 Chinese AI firms of industrial-scale distillation of US models — rohanpaul_ai · 2026-09-09
• [source] CISA advisory: China-based AI firms conduct industrial-scale distillation against US companies — rohanpaul_ai · 2026-09-09
• US agencies accuse 6 Chinese AI firms of industrial-scale distillation; Jensen Huang defends distillation — rohanpaul_ai · 2026-09-09
• NSA, FBI, CISA Report: Nearly Every Major Chinese AI Lab Distills US Models — chrisrohlf · 2026-09-09
• FBI accuses Chinese AI firms of distilling Claude, GPT, Gemini and Grok — ChrisUniverse · 2026-09-10
• US Officials Accuse DeepSeek, Moonshot and Alibaba of Industrial-Scale Distillation — VraserX · 2026-09-11
• NSA, CISA, FBI name six Chinese AI firms accused of copying US frontier models — DynamicWebPaige · 2026-09-11

1 near-duplicate retellings: castrotech