On Thursday, the White House issued a warning that foreign entities are allegedly conducting ‘industrial-scale’ efforts to replicate American artificial intelligence models. These tactics reportedly include jailbreaking and employing extensive networks of fake accounts to access proprietary information and mimic their performance.
In a memorandum titled “Adversarial Distillation of American AI Models,” Michael Kratsios, Assistant to the President for Science and Technology and Director of the Office of Science and Technology Policy, revealed that the U.S. government possesses evidence pointing to coordinated actions aimed at distilling advanced U.S. AI systems.
“We have information suggesting that foreign entities, particularly in China, are orchestrating industrial-scale campaigns to appropriate American AI,” Kratsios stated on X. “Actions will be taken to safeguard American innovation.”
The Trump administration indicated that these operations utilize “tens of thousands of proxy accounts” to bypass detection and employ jailbreaking techniques for systematic capability extraction through what is termed a distillation attack.
“The U.S. has evidence that foreign entities, primarily in China, are conducting industrial-scale distillation campaigns to steal American AI. Protective measures will be implemented,” Kratsios tweeted on April 23, 2026.
A distillation attack involves training a smaller AI model using the outputs from a larger one. This issue is increasingly concerning for U.S. AI companies. In February, Anthropic accused Chinese AI labs DeepSeek, Moonshot, and MiniMax of using around 24,000 fraudulent accounts to extract millions of Claude responses to develop competing systems.
While models produced via unauthorized distillation might not fully match the originals in performance, they can appear similar on specific benchmarks at a reduced cost.
The administration cautioned that such attacks could strip away security measures and other controls intended to maintain AI systems as “ideologically neutral and truth-seeking.”
Federal agencies plan to collaborate with U.S. AI firms to enhance protections for cutting-edge models, coordinate with the private sector to devise defenses against large-scale distillation campaigns, and investigate methods to hold foreign entities accountable.
Although the memorandum acknowledged that legitimate distillation can aid in creating more efficient open-source and open-weight models, it emphasized that unauthorized replication of American AI systems is unacceptable.
“There’s no innovation in systematically extracting and copying the innovations of American industry,” the memo stated. The Office of Science and Technology did not immediately reply to a comment request by Decrypt.