OpenAI, Anthropic, and Google Coordinate to Combat Copying of Their Models in China – Reports

OpenAI, Anthropic, and Google, a subsidiary of Alphabet Inc., have initiated a collaborative effort to address the unauthorized utilization of their artificial intelligence models. This information was revealed by Bloomberg, referencing knowledgeable sources in the technology industry.

The primary challenge these companies face is a practice known as "hostile distillation." This term describes a situation where competitors, primarily from China, extract data from AI models developed by OpenAI, Anthropic, and Google to train their own systems. Such actions violate user agreements and jeopardize the intellectual property of these companies.

According to reports, OpenAI, Anthropic, and Google are sharing information through a non-profit organization called the Frontier Model Forum, which they established in partnership with Microsoft Corp. in 2023. The organization aims to identify practices of unauthorized technology use and protect their developments from competitors.

The method of "distillation" involves using a more advanced AI model to train a new model. This allows for the reproduction of the capabilities of the original system at significantly lower costs than developing from scratch. Certain types of distillation are recognized and even encouraged within the AI community, such as when companies create more compact and efficient versions of their models or allow third-party developers to use them for technologies that do not create direct competition. However, problems arise when this process is used to replicate proprietary technologies without proper authorization.

Most AI models developed in China are open-source, meaning that key components of the system are available for free download and can be run on users' own infrastructure. This makes them cheaper to use, creating additional pressure on American companies in the AI sector. American firms expect to recoup their investments in data centers and infrastructure, which are valued in the hundreds of billions of dollars, through paid access to their products, Bloomberg emphasizes.

According to estimates from American officials, AI laboratories in Silicon Valley lose billions of dollars in profits annually due to unauthorized distillation. These losses are attributed to competitors being able to create their models based on data obtained from models developed by American companies without the need to invest in developing their own technologies.

OpenAI has confirmed to reporters that it is participating in information sharing within the framework of the Frontier Model Forum. In February of this year, the company accused the Chinese firm Deep Seek of attempting to "freely benefit from the developments of OpenAI and other leading American companies in the field of artificial intelligence." This accusation underscores the seriousness of the situation and the importance of collaborative efforts to protect intellectual property amid global competition in the AI market.