The U.S. government says Chinese AI firms are running a systematic campaign to distill capabilities from frontier U.S. models, using billions of tokens and millions of exchanges since late 2024. The advisory names DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun, and Z.AI, and says the companies used the resulting data to strengthen domestic models while trying to avoid detection. #DeepSeek #MoonshotAI #Alibaba #MiniMax #StepFun #ZAI #Claude #ChatGPT #Gemini #Grok
Keypoints
- U.S. agencies accuse Chinese AI firms of industrial-scale model distillation.
- The advisory says the activity has been ongoing since at least late 2024.
- DeepSeek allegedly distilled Claude, ChatGPT, Gemini, and Grok to train R1 and R3.
- Moonshot AI reportedly used millions of queries to train Kimi-K2 and Kimi K3.
- The firms are said to use obfuscation tactics like proxies, multiple accounts, and third-party aggregators.
Read More: https://cyberscoop.com/us-accuses-chinese-ai-companies-distillation/