The White House Office of Science and Technology Policy has warned of deliberate Chinese campaigns to extract and clone US frontier AI models, Ars Technica reports citing a Financial Times story. A memo from OSTP director Michael Kratsios, which the FT reviewed, warns that “foreign entities, principally based in China, are engaged in deliberate, industrial-scale campaigns to distill US frontier AI systems.” China’s foreign ministry, according to the Ars Technica report, called the characterization “slander.”
The memo says that US firms will soon gain access to government intelligence to help them defend against the alleged attacks, according to the Ars Technica report.
What distillation means and who has complained
The technique at the center of the dispute is called distillation — using large numbers of queries to a frontier model to generate training data for a cheaper copycat. The practice received wide attention after the launch of DeepSeek, which OpenAI claimed was trained using outputs from its models.
Since then, multiple US labs have made similar allegations, according to the Ars Technica report. Google claimed in January that “commercially motivated” actors — not limited to China — attempted to clone its Gemini chatbot by querying the model more than 100,000 times to train cheaper alternatives. In February, Anthropic accused Chinese firms DeepSeek, Moonshot, and MiniMax of generating “over 16 million exchanges with Claude through approximately 24,000 fraudulent accounts.” Also in February, OpenAI confirmed that most attacks it observed originated from China.
What the Kratsios memo says
The memo goes beyond the individual lab complaints and frames the activity as a government-level concern. According to the Ars Technica report, Kratsios wrote that Chinese campaigns were “leveraging tens of thousands of proxy accounts to evade detection and using jailbreaking techniques to expose proprietary information.” The memo confirms the US is exploring measures “to hold foreign actors accountable for industrial-scale distillation campaigns.”
AI firms have argued these attacks violate their terms of service. The memo signals the US government is prepared to treat the conduct as something more serious — potentially criminal — though the exact legal mechanisms are not defined in the memo, according to the report.
Congressional direction
The House Select Committee on China issued an April report recommending that Congress direct the Commerce Department’s Bureau of Industry and Security and the Department of Justice to “treat model extraction as industrial espionage” and “impose penalties severe enough to deter Beijing’s theft of American innovation.” How quickly lawmakers would act on that direction is, according to the Ars Technica report, still unclear.
Treating distillation as industrial espionage rather than a terms-of-service violation would change the enforcement tools available: criminal penalties, export controls, and diplomatic pressure would all become relevant, and the legal framework could extend beyond China to any country engaged in similar activity, according to the report.
This article is based on a single source — Ars Technica’s report, itself citing the Financial Times, which reviewed the memo. The underlying Kratsios memo has not been independently cited by a second source in this article.