The White House Office of Science and Technology Policy (OSTP) issued a memo on Thursday accusing China of engaging in widespread theft of artificial intelligence (AI) technology from American companies.
The memo indicates that the U.S. government has received information suggesting that foreign entities, mainly in China, are involved in a deliberate campaign to extract America’s advanced AI systems. OSTP Director Michael Kratsios shared these insights during a recent address, highlighting concerns about “industrial-scale” activities.
“Model Distillation,” is described as a process where output from existing large AI systems is utilized to create smaller versions, almost like a cuckoo bird using other birds’ nests. This technique can be legal or illegal, depending on how it’s implemented.
China’s DeepSeek has become a well-known example, purportedly shaking up the emerging AI market by providing advanced AI capabilities at lower costs. However, critics argue that DeepSeek’s approach veers into outright theft, as it reportedly uses data from competing systems like OpenAI without consent.
The discussion surrounding distillation mirrors debates over “fair use” in copyright. Extracts from publicly accessible AI output, especially with permission, are generally acceptable. However, large-scale distillation performed without consent resembles copyright infringement. It’s akin to a Chinese company announcing it would publish books about a boy wizard with a unique scar.
Kratsios pointed out that Chinese developers are using numerous proxy accounts to avoid detection and employing jailbreak techniques to reveal sensitive information. Jailbreaking entails breaching security measures of software or devices to utilize them in unintended ways.
He further stated, “These coordinated campaigns systematically extract features from American AI models and leverage American expertise and innovation.”
According to the White House memo, the Trump administration plans to “hold foreign actors accountable for their industrial-scale distillation activities.”
In response, the Chinese embassy in Washington dismissed the allegations as “pure defamation.” The embassy emphasized China’s commitment to promoting scientific advancement through cooperation and competition while valuing intellectual property protection.
A representative from the Chinese embassy stated that China is evolving into “not only the world’s factory but also the world’s innovation lab.”
In contrast to China’s dismissive attitude, industry insiders highlight serious concerns regarding these illegal activities. Experts at Anthropic, a U.S. company, recently noted that not only does China steal technology, but it also disables security protocols in “jailbreak” efforts, making the stolen AI even more hazardous. They warn that China’s imitation AI may take on tasks that the original systems were configured to avoid, like developing biological weapons.
Anthropic characterized China’s distillation efforts as an “attack” that breaches service terms.In February, they accused three Chinese projects, including DeepSeek, of illegally extracting data from its Claude AI platform.
“These campaigns are intensifying and becoming more sophisticated. The risk is broader than just one company. To tackle this, rapid and coordinated action among industry, policymakers, and the global AI community is crucial,” said Anthropic, emphasizing that these attacks often contravene export controls.
The Anthropic team expressed concerns that the swift progress from these opportunistic labs could be misinterpreted as proof that export controls are ineffective.
Experts revealed that China utilized a massive network of fake accounts operated through proxies to target Claude, complicating the process of tracing which AI requests were part of the distillation efforts.
One instance showed a single proxy network managing over 20,000 fraudulent accounts simultaneously, blending distilled requests with unrelated customer queries, which made real detection quite challenging.
OpenAI echoed similar accusations regarding the actions of the PRC in February. OpenAI informed the U.S. House Committee on the Communist Party of China that they enhanced their defenses following the launch of DeepSeek, although Chinese attacks on their models persist.
“OpenAI believes the best defense is offense. Investing in U.S. AI leadership and ensuring the global adoption of responsibly developed democratic AI is essential to counteract the swift approach of China’s authoritarian AI,” the company stated.
Reports suggested that Kratsios’ memo might be paving the way for restrictions on sales of advanced AI chips to China.
The Trump administration had tentatively allowed such sales in January, but recent indications pointed out that no shipments occurred yet, according to Commerce Secretary Howard Lutnick.
