OpenAI and Anthropic Accuse China of Massive AI Model Theft Using Refinement Technique

Leading artificial intelligence companies OpenAI and Anthropic have accused China of conducting large-scale technological theft of advanced language models. According to a report by the New York Post, Chinese competitors such as Alibaba allegedly used nearly 25,000 fake accounts to generate millions of outputs from the chatbot "Claude." This data was then reportedly used to train less advanced Chinese models through a process known in the industry as "refinement."

The refinement method is described by experts as the ultimate form of industrial homework theft, enabling foreign entities to replicate the capabilities of expensive AI models at a fraction of the time and cost. Instead of investing billions in independent development, Chinese firms leverage American companies' outputs to build their own models, circumventing U.S. export restrictions on advanced chips.

U.S. government officials are deeply concerned about the effectiveness of this method. Sources close to Anthropic estimate that without refinement, China would lag behind the U.S. by at least 18 months in AI development. Thanks to targeted data theft, this gap has narrowed to just 6 to 9 months, allowing Chinese companies to offer comparable capabilities at much lower prices and gain global market share.

The Trump administration identified this industrial espionage as a major strategic challenge. The White House Office of Science and Technology Policy issued a memo accusing China of large-scale theft campaigns. There is significant worry that stolen technology could be used by Beijing to develop advanced cyberattack capabilities or lethal autonomous weapons powered by AI originally created in Silicon Valley.

The U.S. House Select Committee on China is now pushing for stricter legislation, such as the "AI Oversight Act," which would require government approval for all chip sales or cloud services to concerning countries. Committee Chair John Mulaney emphasized that China’s efforts are driven by theft rather than innovation and stressed the need to close cloud computing vulnerabilities that allow China to exploit American resources in the digital arms race.