White House Accuses China of 'Industrial-Scale' Theft From American AI Models

In brief

• The White House says foreign entities, mainly in China, are running "industrial-scale" campaigns to copy U.S. AI models.
• Officials say attackers use proxy accounts and jailbreaks to extract model capabilities and proprietary data.
• The memo calls for federal agencies and private industry to develop defenses and hold foreign actors accountable.

The White House warned Thursday that “foreign entities” are allegedly carrying out “industrial-scale” campaigns to copy the capabilities of American-based artificial intelligence models, using tactics including jailbreaking and networks of fake accounts to extract proprietary information and replicate their performance.

In a memorandum titled “Adversarial Distillation of American AI Models,” Assistant to the President for Science and Technology and Director of the Office of Science and Technology Policy Michael Kratsios said the U.S. government has information indicating coordinated efforts to distill U.S. frontier AI systems.

“The U.S. has evidence that foreign entities, primarily in China, are running industrial-scale distillation campaigns to steal American AI,” Kratsios wrote on X. “We will be taking action to protect American innovation.”

According to President Donald Trump’s administration, the campaigns are using “tens of thousands of proxy accounts” to evade detection and exploit jailbreak techniques to systematically extract capabilities, in what is known as a distillation attack.

The U.S. has evidence that foreign entities, primarily in China, are running industrial-scale distillation campaigns to steal American AI. We will be taking action to protect American innovation.

These foreign entities are using tens of thousands of proxies and jailbreaking… pic.twitter.com/kSp1FReI7J

— Director Michael Kratsios (@mkratsios47) April 23, 2026

A distillation attack is a method of training a smaller AI model to learn from the outputs of a larger one. The issue has become a growing concern among U.S. AI companies. In February, Anthropic accused Chinese AI labs DeepSeek, Moonshot, and MiniMax of extracting millions of Claude responses—using roughly 24,000 fraudulent accounts—to train competing systems.

Models developed through unauthorized distillation campaigns may not match the full performance of the originals. Still, they can appear comparable on select benchmarks at a fraction of the cost.

The administration warned that distillation attacks could also remove security safeguards and other controls designed to keep AI systems “ideologically neutral and truth-seeking.”

The Trump administration said federal agencies will work with U.S. AI companies to strengthen protections around frontier models, coordinate with private industry to develop defenses against large-scale distillation campaigns, and explore ways to hold foreign actors accountable.

While the memo acknowledged that lawful distillation can help create smaller, more efficient open-source and open-weight models, it said unauthorized efforts to copy American AI systems cross the line.

“There is nothing innovative about systematically extracting and copying the innovations of American industry,” the memo said.

The Office of Science and Technology did not immediately respond to a request for comment by Decrypt.

Daily Debrief Newsletter