NSA, CISA, and FBI say six China-based AI firms ran industrial-scale campaigns to distill Claude, GPT, Gemini, and Grok into cheaper rival models.
Read the full story: NSA, CISA, FBI Accuse 6 Chinese AI Labs of Model Distillation →
Transcript
Three US security agencies just went public with something big. The NSA, CISA, and FBI named six Chinese AI companies, DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun, and Z.AI, accusing them of industrial scale distillation. That means pulling billions of tokens out of Claude, GPT, Gemini, and Grok to train cheaper copies. DeepSeek allegedly used Claude, GPT, and Gemini to build R1 and V3. Moonshot allegedly trained Kimi K3 on Claude. Alibaba allegedly distilled Claude 4 and GPT-5 last year.
The methods described: fake accounts, bulk subscriptions, proxy relays to dodge detection. This is bigger than July's single Moonshot sanctions case. It's three federal agencies, six companies, one formal advisory. If the allegations hold, it says the AI moat everyone's spent billions building might be a lot thinner than it looks.