U.S. Accuses Chinese AI Firms of Industrial-Scale Model Theft

The United States government has accused several Chinese artificial intelligence firms of conducting industrial-scale campaigns to steal intellectual property from American AI laboratories. Through a White House memorandum and a subsequent global diplomatic cable from the State Department, officials alleged that companies including DeepSeek, Moonshot AI, and MiniMax utilized a process called distillation. This technique involves using proxy accounts and jailbreaking to harvest outputs from frontier U.S. systems to train smaller, cheaper imitation models that bypass security protocols and neutrality mechanisms.

Michael Kratsios, the White House Chief Science and Technology Adviser, stated that these activities allow foreign actors to commercialize derivative systems without the cost of original development. American AI firms OpenAI and Anthropic supported these claims, with Anthropic reporting the use of over 20,000 fraudulent accounts to extract data from its Claude system. In response, the U.S. House Foreign Affairs Committee unanimously passed a bill to identify and sanction foreign entities engaging in such extractions.

The Government of China and its embassy in Washington rejected the allegations as groundless and a deliberate attempt to suppress China's technological progress. DeepSeek denied using synthetic data from OpenAI, claiming its models rely on web-crawled data, and subsequently launched a preview of its V4 model adapted for Huawei chip technology. These tensions have escalated ahead of a scheduled meeting in Beijing between President Donald Trump and President Xi Jinping.