U.S. agencies say several Chinese AI companies systematically pulled capabilities and outputs from leading American AI models to improve their own systems. The report names DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun, and Z.AI, and says they extracted billions of tokens across millions of requests from Claude, GPT, Gemini, and Grok since at least late 2024 using large-scale distillation, regional restriction evasion, centralized request routing, metadata sanitization, and quota and cost optimization.
Why it matters: This matters to AI vendors, cloud providers, and enterprise users because it describes an ongoing, well-resourced campaign to siphon valuable model capabilities rather than a one-off abuse case. Organizations operating frontier models or AI APIs should review the agencies’ detection and mitigation guidance now, especially around abuse monitoring, access controls, and coordinated infrastructure-level defenses.
Kevin Townsend
2026.09.09
100% relevant
This article establishes a distinct U.S. government-attributed campaign focused on systematic AI model distillation by named Chinese firms, rather than a patch, single intrusion, or previously tracked influence or espionage case.
← Back to all stories