Khanna Seeks Data From AI Labs on Alleged Chinese Theft of Model Weights
The ranking Democrat on the House Select Committee on the Chinese Communist Party has written to four major American artificial intelligence companies, seeking information about alleged efforts by Beijing to obtain model weights from their systems.
Rep. Ro Khanna (D-Calif.) sent the requests on Thursday to OpenAI, Anthropic, Meta and Alphabet, the parent company of Google. He said the theft of such technology could "erode America's AI lead with the stroke of a keyboard."
Model weights are the numerical parameters within AI systems that hold the knowledge a machine has learned and shape what it produces.
In letters obtained by The Hill, Khanna warned that if the People's Republic of China were to steal a proprietary model weight from a leading U.S. frontier laboratory through state-sponsored hacking, the CCP would immediately skip past years of costly research and gain access to advanced capabilities that American policy has long aimed to keep from Beijing.
Earlier this year, Michael Kratsios, director of the White House Office of Science and Technology, said in a memo that Washington had learned Chinese entities were using thousands of proxy accounts and jailbreaking methods to reach proprietary information.
That practice is known as distillation, in which the outputs of a more powerful system are used to train weaker ones. The resulting models do not match the full abilities of the original, but they let foreign actors put out products that look comparable at a lower price.
Washington has moved to restrict the flow of American technology to China amid a close AI competition. Late last year, President Trump faced criticism from members of his own party after deciding to permit Nvidia to sell its H200 chips to China.
Reuters was first to report the existence of Khanna's letters.
The California Democrat said that even without stealing a complete model weight, Chinese actors could still extract "value [intellectual property]" from the American companies' systems and reach sensitive research.
Khanna asked the firms to provide a list of every known case over the past two years in which China or another hostile actor tried to obtain model weights and intellectual property, as well as details of the companies' efforts to strengthen cybersecurity.
He wrote that just as the companies are responsible for guarding against dangerous uses of their models and the effects of AI on humanity and society, they also have a duty to ensure that a cybersecurity breach at their firm does not put American national security at risk.
The companies and the Chinese Embassy in Washington did not immediately respond to requests for comment.
to like, bookmark, and comment.
Comments
Join the conversation:
No comments yet. Be the first to share your thoughts.