OpenAI, Anthropic, Google team up to stop Chinese AI distillation threat

In a significant alliance, three of the leading names in artificial intelligence—OpenAI, Anthropic, and Google—are collaborating to thwart attempts by foreign entities, particularly from China, to replicate their advanced AI technologies. This initiative, reported by Bloomberg, is facilitated through the Frontier Model Forum, an industry group that also includes Microsoft, which was established in 2023. The primary goal of this collaboration is to identify and prevent what is known as 'adversarial distillation.' This technique involves unauthorized parties attempting to clone AI systems, undermining the extensive investments made by these companies. Distillation itself is a legitimate process in AI development where a robust 'teacher' model trains a smaller 'student' model, allowing companies to optimize their efficiency. However, when used by external developers without consent, it translates into a form of intellectual theft, costing American firms billions in potential revenue each year, according to U.S. officials. The urgency of this cooperation stems from a startling event in January 2025, when the Chinese startup DeepSeek launched its R1 AI model, which rivaled the performance of top U.S. systems at a significantly lower cost. This revelation sent shockwaves through the tech world, erasing hundreds of billions in market value for American AI companies. In response, Microsoft and OpenAI began investigations into whether DeepSeek had illicitly extracted data from OpenAI's models to create R1. By February 2025, OpenAI had alerted U.S. lawmakers about DeepSeek's ongoing efforts to exploit vulnerabilities in its systems. A memo addressed to the House Select Committee on China alleged that DeepSeek was attempting to benefit from the technological advancements developed by OpenAI and other U.S. research institutions. Anthropic has been particularly vocal, having previously barred Chinese-controlled firms from utilizing its Claude AI chatbot. Earlier this year, it named three Chinese companies—DeepSeek, Moonshot, and MiniMax—as culprits in illegally mimicking Claude's functionalities through distillation. The company has expressed concerns that this issue poses broader national security risks, especially as copied AI systems often lack the essential safety measures designed to prevent misuse. Meanwhile, Google has reported a notable increase in attempts to appropriate its AI capabilities. Despite the seriousness of the threat, the sharing of intelligence among these companies faces challenges due to uncertainties regarding legal limitations under current antitrust laws. There is a clear need for more regulatory guidance from the U.S. government, as indicated by sources familiar with the situation. The Trump administration had previously shown support for formalizing such collaborative efforts, proposing an information-sharing center as part of its AI Action Plan, aimed at countering adversarial distillation.