OpenAI, Google, and Anthropic Unite Against Chinese Distillation

A Strategic Alliance to Protect Innovation

OpenAI, Anthropic, and Google have decided to join forces to counter the growing threat of unauthorized copying of their artificial intelligence models by Chinese companies. This collaboration, reported by Bloomberg, is part of the "Frontier Model Forum," an information-sharing space established in 2023.

The Threat of Adversarial Distillation

Adversarial distillation is at the heart of these companies' concerns. This process involves using the outputs of an existing AI model to train a copycat model, often at a lower cost. Stanford's Alpaca model was one of the first to demonstrate the feasibility of this approach, which has become a major issue for American companies.

Colossal Financial Losses

U.S. authorities estimate that this practice costs AI labs in the United States billions of dollars in lost revenue each year. In February, OpenAI had already alerted Congress about the sophisticated methods employed by Deepseek to extract data from American models. Anthropic has also identified Deepseek, Moonshot, and Minimax as key players in this practice.

A Model Inspired by Cybersecurity

This collaboration between OpenAI, Anthropic, and Google draws inspiration from the cybersecurity sector, where information sharing about attacks is a common practice. By uniting their efforts, these companies hope to better detect and counter attempts at adversarial distillation, thereby protecting their innovations and revenues.