Key facts
- U.S. security agencies accused Chinese AI companies of "aggressive, industrial-scale distillation activities."
- The U.S. State Department issued a global warning alleging systematic theft of American AI technology by Chinese firms.
- The alleged theft involves model architectures and training methodologies.
- DeepSeek, Moonshot, and MiniMax are among the Chinese firms accused of using distillation tactics.
- OpenAI confirmed that most attacks it observed originated from China.
U.S. cyber and law enforcement officials have accused Chinese artificial intelligence companies of engaging in "aggressive, industrial-scale distillation activities," according to a statement from three U.S. security agencies. Distillation is a process where smaller AI models are trained using the output from larger, more expensive models to reduce training costs.
The U.S. State Department has issued a global warning alleging that Chinese companies, including DeepSeek, are systematically stealing American AI technology. This coordinated campaign is seen by the administration as an effort to undermine U.S. innovation and intellectual property.
Reports indicate that the White House has described the actions as "mass AI theft," with accusations that China is copying American AI models on an "industrial-scale." The alleged theft reportedly involves not only software but also underlying model architectures and training methodologies developed at significant cost by U.S. companies.
DeepSeek, a Hangzhou-based AI company, has come under particular scrutiny. The company claims its V3 model was trained for approximately $6 million, a fraction of the cost for OpenAI's GPT-4, and used significantly less computing power than Meta's Llama 3.1. While DeepSeek attributes these gains to architectural innovations and efficient training techniques, U.S. officials suggest unauthorized access to American AI research may be involved. No public evidence has been presented to substantiate these specific allegations.
Google claimed in January that "commercially motivated" actors attempted to clone its Gemini AI chatbot. In February, Anthropic accused Chinese firms DeepSeek, Moonshot, and MiniMax of using distillation to generate millions of exchanges with its Claude model through fraudulent accounts. OpenAI confirmed that most attacks it observed originated from China.
U.S. officials warn that these distillation attacks threaten to help China quickly advance in the AI race, potentially discouraging investment in costly research and development if competitors can replicate outcomes without bearing the initial burden. Critics, however, caution against conflating legitimate technological catch-up with illicit theft, pointing to the global nature of AI research and the possibility of algorithmic efficiencies explaining performance gains.