The U.S. Plans to Allow AI Giants to Legally Form Cartels to Block Chinese Model "Distillation"

According to InferenSeek monitoring, bipartisan U.S. lawmakers have introduced the "CATS Act." The bill aims to provide limited antitrust protection for AI companies. Competitors can share intelligence on model theft, distillation, and cyber attacks. In the event of a national security risk, they can also collectively delay or restrict the deployment of high-risk models.

Distillation, as it is called, involves extensively querying a strong model and collecting responses to then train a cheaper model. The proponents have singled out China as a primary target. Previously, Anthropic, OpenAI, and Google have all accused Chinese labs of extensively extracting their model capabilities. The U.S. government has recently also named the dark side of the Moon's Kimi K3.

Before jointly restricting a model, companies must provide written notice to the U.S. Department of Justice. They must also demonstrate that their actions are solely aimed at addressing security risks. Price fixing, market allocation, monopolies, and group boycotts remain illegal.