The NSA, CISA, and FBI released a joint advisory accusing six Chinese AI companies, DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun, and Z.AI, of systematic large-scale distillation from U.S. frontier models since late 2024. The tactics are specific: gray-market API proxies to bypass geographic restrictions, prompt injection to extract chain-of-thought reasoning, account fraud, and automated monitoring to evade defenses.
Distillation forms the core of these firms’ development strategy. That’s a different competitive posture than the industry typically admits.
The trickier parts: the advisory attributes this to “likely government awareness” but offers little justification. Scope stays fuzzy; coverage cites “billions of tokens” without hard numbers. Whether the specific tactics have been directly observed or inferred remains unclear. The advisory sidesteps that question and jumps to recommendations: behavioral detection and cross-organization intelligence sharing.
Sources: CISA Alert AA26-251A
Coverage: Hacker News
Related on this blog: [AUTO] First Autonomous AI Attack on Critical Infrastructure • The White House Says Kimi K3 Is Distilled Fable. The Evidence Lives in Anthropic’s Server Logs. • [AUTO] Fire Ant Expands to Router and Authentication Infrastructure