China-linked AI firms face US scrutiny over model theft

The President Trump administration has announced plans to fight what it called “industrial-scale campaigns” aimed at copying artificial intelligence technology from American companies. 

The White House Office of Science and Technology Policy said foreign entities are targeting major US AI firms through unauthorized model distillation.

Michael J. Kratsios, assistant to the president for the White House science office, said the government has “information” that foreign entities, mainly based in China, are trying to extract capabilities from US AI models. The statement said these groups use proxy accounts and jailbreaking methods to avoid detection.

Kratsios said, “Models developed from surreptitious, unauthorized distillation campaigns like this do not replicate the full performance of the original.” He added that such models can still allow foreign actors to release products that appear close to US systems on some benchmarks at a much lower cost.

US says proxy accounts helped hide activity

The White House said some foreign companies used “tens of thousands of proxy accounts” to mask their activity while probing American AI models. It also said the groups used jailbreaking techniques to expose private or protected model information.

According to the science office, these campaigns aim to extract useful features from American models without permission. The statement said, “These coordinated campaigns systematically extract capabilities from American AI models, exploiting American expertise and innovation.”

The White House also said models built through these methods may lack security controls. It warned that copied systems could move away from being “neutral and truth-seeking” if safety protections are removed or weakened.

Moreover, the statement follows claims made by Anthropic in late February. The Claude developer accused three Chinese AI firms, DeepSeek, Moonshot, and MiniMax, of carrying out distillation attacks against its models.

Anthropic said the firms created more than 16 million exchanges with its AI models through about 24,000 “fraudulent accounts.” The company said the activity targeted capabilities such as coding, agentic reasoning, data analysis, grading tasks, and computer vision.

The case has added focus to how frontier AI companies protect model access. AI firms charge users through token-based pricing, and lower-cost competitors can gain market attention by offering similar performance on selected tasks.

US plans closer work with private AI firms

The White House science office said the administration will work with US companies to share information about large attacks. It also plans to help the private sector coordinate stronger defenses against foreign actors.

The administration said it will explore measures to “hold foreign actors accountable.” It did not provide a detailed list of possible penalties or enforcement steps in the statement.

The move comes as AI competition between the United States and China continues to grow. US officials have framed advanced AI systems as a core technology for national security, business productivity, and future economic power.