Leading artificial intelligence companies Anthropic and OpenAI have accused Chinese firms of engaging in a significant technology theft operation, using a method called "distillation" to illegally replicate advanced AI models. Experts warn this practice not only threatens the United States' narrow lead in artificial intelligence but also poses a risk to national security.
On June 24, Anthropic sent a letter to Congress directly accusing Alibaba of "brazen" AI theft. This accusation followed a similar claim in February by OpenAI, which stated that China-based DeepSeek was "free-riding" on its work. Google also reported a surge in "malicious activity" originating from China, Iran, North Korea, and Russia around the same time.
These alleged theft campaigns have reportedly enabled Chinese companies to offer AI services at significantly lower prices than their U.S. counterparts, attracting global customers. Six of the top 10 most-used AI models are now developed by Chinese tech firms, including Z.AI, Minimax, and DeepSeek, which reportedly possess capabilities approaching those of Anthropic's Claude and OpenAI's ChatGPT.
Sources close to Anthropic suggest that without distillation, China would be approximately 18 months behind U.S. frontier models, rather than the current estimated 6-9 month gap. Anthropic has reportedly briefed Trump administration national security officials on these distillation attacks.
Distillation and its Implications
Distillation is a legitimate technique used by AI labs to train smaller, more efficient models using larger, more capable ones. However, the practice becomes illicit when it is used without permission to replicate a competitor's technology. Theresa Payton, former White House chief information officer, described these actions as "the ultimate form of industrial homework theft," allowing foreign adversaries to clone advanced AI models at a fraction of the cost and effort.
The strategy is particularly crucial for China, as U.S. export controls on advanced computer chips, such as those produced by Nvidia, have hindered its ability to develop sophisticated AI models domestically. Distillation allows Chinese labs to bypass these restrictions by leveraging existing U.S. AI infrastructure.
Anthropic's letter to the Senate Banking Committee detailed evidence of Alibaba employees allegedly generating millions of outputs from Anthropic's Claude chatbot using thousands of fraudulent accounts. While Alibaba subsequently banned its employees from using Claude, Anthropic noted that China-based entities are already prohibited from using its models.
OpenAI has also briefed congressional staff on the misuse of AI by China, including its role in malicious influence operations. The company advocates for stronger policy tools to hold malicious actors accountable and suggests the NSA's AI Security Center could act as a central hub for the tech industry to share information on such threats.
National Security Concerns and Legislative Efforts
The House Select Committee on China has highlighted China's history of AI chip smuggling and model distillation, pushing for legislation to counter these threats. Potential measures include the AI Overwatch Act, which would require government permission for advanced AI chip sales to "countries of concern."
Select Committee Chairman John Moolenaar stated that China's AI development is driven by theft rather than innovation, posing a threat to U.S. tech companies and national security. He emphasized the need to prevent China's actions in the AI arms race and to close loopholes in cloud computing that China exploits.
Jared Dunnmon, a former technical director for AI at the Pentagon, noted the difficulty in quantifying the extent to which Chinese AI performance gains are due to distillation versus genuine innovation. He warned that these attacks not only threaten U.S. AI leadership but also risk enabling adversaries to develop advanced hacking or autonomous weapon capabilities. The concern is that China could get close enough to U.S. capabilities to offer competitive AI services at much lower costs, potentially leading global developers to build on Chinese AI software stacks.