Leading artificial intelligence developers OpenAI, Anthropic, and Google have reportedly teamed up through the Frontier Model Forum to counter efforts by Chinese companies to copy their advanced AI models. This collaboration aims to address concerns over the extraction of capabilities from cutting-edge U.S. AI models, which could provide Chinese competitors with an unfair advantage in the global AI race.
The firms are sharing information via the Frontier Model Forum, an industry nonprofit co-founded by OpenAI, Anthropic, Google, and Microsoft in 2023. The primary goal of this information exchange is to detect “adversarial distillation attempts” that violate their terms of service. This concerted effort highlights the significant concern among U.S. AI companies regarding such practices.
Recent allegations underscore the urgency of this collaboration. Anthropic has accused three prominent Chinese AI organizations—DeepSeek, Moonshot AI, and MiniMax—of engaging in a sophisticated distillation process. According to Anthropic, these groups leveraged approximately 16 million interactions across 24,000 fraudulent accounts to extract knowledge and capabilities from its Claude model. Similarly, OpenAI reportedly sent a letter to U.S. lawmakers alleging that DeepSeek had trained its models on outputs derived from OpenAI’s proprietary models. Google’s threat intelligence group has also issued warnings about an increase in distillation attacks targeting its Gemini models.
Adversarial distillation is a technique that enables smaller models to learn efficiently from larger, more advanced ones, potentially achieving 80% to 90% of a frontier model’s performance with considerably less computing power. This method can drastically reduce operating costs and accelerate price competition, contributing to the perceived threat to original innovators.
The Frontier Model Forum’s broader mission includes advancing AI safety research, identifying best practices, and facilitating information sharing among various stakeholders to ensure the safe and responsible development of frontier AI models. In March 2025, member firms signed an agreement to share information about vulnerabilities, threats, and capabilities of concern related to frontier AI, encompassing exploitable flaws, unauthorized access attempts, and capabilities that could cause large-scale harm.
The differing approaches to intellectual property protection for AI-generated content between the U.S. and China further complicate the landscape. While a Chinese court has ruled in favor of copyright protection for an AI-generated image, the U.S. Copyright Office generally maintains that AI-generated works lack human authorship and are therefore not protectable.