OpenAI, Anthropic, and Google have begun sharing intelligence through the Frontier Model Forum to combat unauthorized AI model copying by Chinese competitors. The three rivals are coordinating to detect "adversarial distillation" attacks, where foreign developers systematically query systems like ChatGPT, Claude, and Gemini to extract outputs used to train cheaper competing models. U.S. officials estimate these efforts cost Silicon Valley labs billions of dollars annually.
The collaboration intensified after DeepSeek released its R1 reasoning model in January 2025, which matched leading American systems at a fraction of the cost. OpenAI told lawmakers that DeepSeek tried to "free-ride on the capabilities developed by OpenAI and other U.S. frontier labs." Anthropic has specifically accused DeepSeek, Moonshot, and MiniMax of extracting Claude's capabilities and has banned Chinese-controlled companies from its platform entirely.
Enforcement measures include detecting abnormal traffic patterns, canceling accounts, blocking IP ranges, and modifying rate limits. The Trump administration has signaled support for formalizing this collaboration, including establishing a dedicated information-sharing center for AI security threats.
