Technology
U.S. agencies allege Chinese firms systematically mined American AI models
An advisory from intelligence and security officials names six developers accused of illicit model distillation campaigns since 2024.
Latest update: U.S. agencies named six Chinese developers, including DeepSeek, Alibaba, and Moonshot AI, accused of distilling American artificial intelligence models.
The short version
- The FBI, NSA, and CISA accused six Chinese AI firms of running large-scale distillation campaigns to extract data from American models like ChatGPT, Claude, Gemini, and Grok.[NBC News · Slashdot]
- Named developers include DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun, and Z.AI, which allegedly used the queries to train their own systems.[NBC News · Slashdot]
- Federal officials stated the activity likely occurred with Chinese government knowledge, though they stopped short of alleging direct intelligence agency involvement.[NBC News]
- A spokesperson for the Chinese embassy in Washington did not immediately respond to requests for comment.[NBC News]
Key facts
- The NSA, CISA, and FBI published an alert accusing Chinese AI firms of conducting industrial-scale distillation campaigns against leading U.S. models since at least 2024.[NBC News · Slashdot]
- The advisory identified six Chinese developers: DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun, and Z.AI.[NBC News]
- DeepSeek allegedly distilled models including Claude, Gemini, ChatGPT, and Grok to create synthetic training data for its R1 and R3 systems.[Slashdot]
- Moonshot AI allegedly extracted capabilities from 18 U.S. models, including Anthropic's Fable 5, to develop its Kimi-K2 and Kimi K3 systems.[Slashdot]
What remains uncertain
Sources
Outlet counts describe coverage, not independent confirmation. Reports may share a wire service or original source.