White House memo claims mass AI theft by Chinese firms

The White House announced plans to strengthen collaboration with American AI companies to counter ‘large-scale efforts’ by international actors to siphon technological innovations. In an internal memo, Michael Kratsios, the Director of Science and Technology Policy, highlighted new evidence suggesting that entities based in China are leveraging American firms to access cutting-edge advancements. He noted that these organizations use a technique known as ‘distilling’ to replicate AI technologies developed domestically, effectively bypassing traditional safeguards.

Chinese Embassy Responds to Allegations

A spokesperson from China’s US embassy in Washington DC contested the claim, asserting that the progress made by Chinese firms is due to their own commitment and global partnerships. ‘China is not only the world’s factory but is also becoming the world’s innovation lab,’ the representative emphasized, adding that such development brings shared benefits through cooperative efforts.

“China’s development is the result of its own dedication and effort as well as international cooperation that delivers mutual benefits.”

Kratsios outlined the goal of the strategy as disrupting American research and development, while preventing unauthorized access to confidential data. The memo listed four measures to address the issue, though it did not specify which entities would be targeted. A White House spokesperson confirmed the memo’s content but offered no further details.

Distillation campaigns involve companies managing thousands of user accounts for a single AI tool, mimicking regular activity. These accounts later engage in coordinated actions to extract sensitive information from models, which is then integrated into their own systems. Kratsios warned that as detection methods improve, foreign firms relying on such tactics may doubt the reliability of their models.

Targeted Firms and Ongoing Concerns

Although Kratsios did not name specific entities, major AI firms have acknowledged the threat. Anthropic recently identified three Chinese laboratories—DeepSeek, Moonshot, and MiniMax—as conducting distillation attacks to replicate their models. OpenAI has similarly accused DeepSeek of copying its technology. The three labs have not yet commented on the claims.

DeepSeek, launched last year, rapidly gained popularity among users. The company claimed its model required only a few million dollars to develop, far less than the hundreds of billions spent by competitors. However, it faced a significant outage last month and is preparing a revised version of its AI model.

US President Donald Trump is set to visit China in May, potentially addressing concerns over intellectual property and technology transfer. The White House’s memo underscores growing worries about the global competition in AI innovation and the role of foreign firms in shaping its trajectory.