U.S. Agencies Accuse Six Chinese AI Firms of Systematically Copying American Models

Summary

The FBI, National Security Agency and Cybersecurity and Infrastructure Security Agency said in a joint alert that six leading Chinese AI developers have systematically mined American AI models since 2024. The companies named are DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun and Z.AI. The alert says the companies used distillation, a technique in which one model’s outputs help develop another, to draw from Anthropic’s Claude, OpenAI’s ChatGPT, Google’s Gemini and xAI’s Grok. Distillation is common across the AI industry, but it can violate model providers’ terms of service. U.S. officials said the scale and sophistication of the alleged campaigns suggest distillation was central to the companies’ model development, rather than a supplementary technique. The report said the activity likely occurred with Chinese government awareness, while stopping short of alleging Chinese intelligence involvement. China’s Foreign Ministry dismissed the accusations as groundless and described the country’s AI progress as the result of scientific and technological self-reliance. The agencies recommended that U.S. AI companies strengthen subscription-user verification, share information about suspicious activity and, in some cases, modify responses to potentially malicious queries. The alert follows earlier public accusations by OpenAI and Google and a White House memorandum identifying Chinese distillation as a critical issue.