NSA, CISA, FBI accuse 6 Chinese AI firms of industrial-scale distillation of US models
rohanpaul_ai · x · 2026-09-09
- A joint NSA/CISA/FBI advisory (AA26-251A) names DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun and Z.AI as running industrial-scale distillation campaigns against US frontier models since at least late 2024.
- Alleged methods: gray-market API proxies ("transfer stations"), bulk premium subscriptions shared across dev teams, and third-party aggregators stripping account metadata.
- Moonshot AI allegedly distilled 17 US models including Anthropic's Claude Fable 5 to train Kimi-K3; DeepSeek's prompts reportedly pushed models to expose hidden chain-of-thought, transferring reasoning method rather than answers.
- The advisory argues DeepSeek's headline training cost only covers compute and omits data cost — allegedly obtained by distilling US models rather than original research.
- Controversial mitigation: serve suspected distillers undisclosed degraded responses; detection indicators (24/7 usage, new accounts at max throughput, cache-hit-optimized traffic) overlap with ordinary enterprise agent fleets, leaving providers to decide who gets silently downgraded.
More from Models
- MissionReport benchmark: only 3 of 540 AI-generated government reports pass all criteria — Exp_Mark · 2026-10-27
- Local beats cloud: Qwen 3.8 FN handles private student data as well as frontier models — dr_lm · 2026-10-27
- Aidan Gomez: model's CoT said no tool calls right before making them — aidangomez · 2026-10-27
- Nat Lambert: Cohere's model is narrow RL training on internal infra ill-suited for open weights — natolambert · 2026-10-27
- OpenAI teases DOTS is coming to Europe 'soon' — mark_k · 2026-10-27
- Clifford Pickover challenges AI to decipher a message he encoded in Norse runes — pickover · 2026-10-27