Overview
- A joint advisory from the NSA, CISA and FBI on September 8 named DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun and Z. AI and said they ran high-volume knowledge‑distillation campaigns against U.S. frontier models since late 2024.
- The agencies described extraction at scale as billions of tokens and millions of requests that targeted specific capabilities such as advanced reasoning, coding, software engineering and chain‑of‑thought outputs.
- Officials listed evasive tactics including pooled accounts, continuous 24‑hour usage, gray‑market proxy “transfer stations,” and routing requests through many APIs and clouds to hide origins.
- As an operational defense the advisory urged providers to strengthen identity checks, share threat information, and, in confirmed cases, alter or limit model responses for suspected abusers to blunt capability extraction.
- China’s commerce and foreign ministries rejected the allegations as baseless and warned of countermeasures, and policymakers and vendors are weighing export controls, sanctions or service changes that could disrupt legitimate high‑volume enterprise workloads.