U.S. Accuses Chinese AI Firms of Systematic Distillation of Proprietary Models

• •Chinese AI firms are accused of systematic extraction of U.S. AI model capabilities.
• •The distillation activities have been ongoing since late 2024 and involve billions of tokens.
• •U.S. agencies recommend immediate actions for AI companies to protect against these threats.

U.S. cybersecurity agencies have accused Chinese AI companies of conducting systematic extraction of proprietary functionalities from leading U.S. AI models through industrial-scale distillation attacks. This activity, which has been ongoing since at least late 2024, involves companies like DeepSeek, Moonshot AI, Alibaba, and MiniMax extracting billions of tokens from models such as Claude, GPT, Gemini, and Grok. The U.S. agencies described these operations as a core part of China's AI development strategy, enabling faster and cheaper advancements in their AI capabilities. The advisory outlines various tactics used by these companies to bypass safeguards, including spreading requests across multiple accounts and using proxies. The advisory urges U.S. AI firms to implement detection and mitigation strategies to safeguard against these attacks.