U.S. national security agencies have accused six Chinese AI companies of conducting industrial-scale data theft to extract proprietary capabilities from leading American frontier models like GPT and Claude.

Key Points

• The NSA, CISA, and FBI issued a joint advisory naming DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun, and Z.AI as participants in unauthorized model distillation.
• Agencies allege these companies used millions of requests and fraudulent accounts to bypass geographic restrictions and extract billions of tokens from U.S. models.
• The activity, occurring since late 2024, reportedly aims to replicate advanced reasoning and optimization capabilities without incurring the high costs of independent research.
• Beijing’s Ministry of Foreign Affairs denied the allegations, asserting that China’s AI progress stems from domestic innovation and international cooperation.
• U.S. officials and industry leaders, including Treasury Secretary Scott Bessent, emphasized that maintaining a lead in AI infrastructure is critical to national security.

Why it Matters

This dispute highlights the intensifying global race for artificial intelligence supremacy and the risks posed by intellectual property theft in the tech sector. If Chinese firms successfully replicate U.S. frontier models through distillation, it could significantly erode the competitive advantage of American companies and impact future military and cyber capabilities.