Overview
- The National Security Agency, CISA and the FBI issued a joint advisory on Sept. 8 that accuses DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun and Z. AI of extracting billions of tokens from U.S. frontier models through millions of requests.
- Agencies say the campaigns used tactics to hide origins such as gray‑market proxy “transfer stations,” pools of fraudulent accounts across many IPs, 24‑hour query patterns and new subscriptions consuming full quotas.
- The advisory tells providers to detect suspicious prompts and accounts, share threat data and in some cases alter or downgrade responses to confirmed distillation requests to reduce the value of stolen outputs.
- Chinese officials have rejected the claims as baseless, accused the U.S. of double standards and warned of countermeasures if Washington uses the advisory to restrict Chinese AI firms.
- The dispute raises near‑term policy risks for sanctions or export controls and puts pressure on AI providers and enterprise customers because stricter identity checks or undisclosed model downgrades could disrupt legitimate high‑volume workloads.