OpenAI, the developer of ChatGPT, warned the U.S. Congress that Chinese competitor DeepSeek is illicitly siphoning outputs from major U.S. artificial intelligence (AI) models to train its own AI model "R1."
According to Bloomberg News, OpenAI claimed in a memo submitted to the House Select Committee on the Strategic Competition between the United States and the Chinese Communist Party that DeepSeek is using a "distillation" technique to extract outputs from U.S. AI models. Distillation is a method of creating a model with similar performance by using answers produced by other AI models as training data.
U.S. AI corporations also use the distillation technique. It can be used to build a lightweight, lower-tier model with capabilities comparable to a top-tier model. Google, for example, builds "Gemini Flash" based on its top-tier model "Gemini Pro."
Bloomberg reported that OpenAI raised concerns immediately after the "R1" model was released last year and launched an investigation with partner Microsoft (MS) to verify whether DeepSeek had obtained data without authorization. OpenAI criticized it as "part of an ongoing attempt to free-ride on a competitor's development capabilities." It also noted that safety features applied to prevent misuse in sensitive areas such as biology and chemistry could be neutralized during the distillation process.
Because Chinese AI models, including DeepSeek, are offered for free, the spread of distillation would inflict major harm on U.S. corporations developing AI models. OpenAI stressed that such unfair practices risk weakening the United States' edge in the U.S.-China AI competition.
OpenAI said it "observed an account associated with a DeepSeek employee developing methods to circumvent access restrictions and hiding the source to access the model," claiming that approaches for unauthorized extraction are becoming more sophisticated and, at times, involved extraction activities linked to Russia.
John Moolenaar, the Republican Chairperson of the House China Committee, said, "Steal, copy, and eliminate competitors is the Chinese Communist Party's typical playbook," adding, "Chinese corporations will continue to extract U.S. AI models and exploit them for their own gain."
Earlier, David Sacks, known as the "AI czar" and chairperson of the White House Office of Science and Technology Policy, also criticized Chinese AI corporations' distillation techniques in media interviews last year. He said, "There is evidence that DeepSeek illicitly extracted data from U.S. AI models such as OpenAI."
OpenAI's warning comes as political concern grows over exports of Nvidia AI chips to China. President Donald Trump eased some semiconductor export restrictions late last year, allowing Nvidia to sell its H200 chip, released two years ago, to China.