OpenAI, Anthropic and Google are sharing threat data. The effort targets Chinese competitors extracting results from US frontier models to train cheap replicas and win customers on price. It marks an unusual truce among direct rivals.
The coordination runs through the Frontier Model Forum. The nonprofit was founded in 2023 by the three labs and Microsoft to discuss frontier-model risks and now shares indicators of adversarial distillation, people familiar said. OpenAI confirmed it is participating and cited a recent memo accusing DeepSeek of trying to “free-ride” on US capabilities.
Rivals OpenAI, Anthropic, and Google have begun working together to try to clamp down on Chinese competitors extracting results from cutting-edge US artificial intelligence models to gain an edge in the global AI race https://t.co/fWdKdmlS2t
— Bloomberg (@business) April 6, 2026
Distillation trains a newer “student” model on a “teacher” model’s answers. Some versions are accepted when companies compress their own systems, or when developers build non-competitive tools under permission. The controversy rises when third parties use it to copy proprietary work, strip safety controls, and ship open-weight alternatives at lower cost.
US officials have estimated unauthorized distillation costs Silicon Valley labs billions in annual profit, a person familiar said. That economic hit lands as US firms try to fund data centers and model training with paid access, while many Chinese models are open weight and cheaper to run. It also adds a national security angle, because copied models can lose guardrails meant to block high-risk misuse such as pathogen support.
Anthropic has already tightened access in response. Last year it blocked Chinese-controlled companies from using Claude, and in February it said DeepSeek, Moonshot and MiniMax were illicitly extracting capability via distillation. Can the labs stop copying without making model access unusable for legitimate developers?
The information-sharing plan mirrors cybersecurity playbooks, where firms swap indicators to spot coordinated abuse faster. Trump’s AI Action Plan called for an information sharing and analysis center, but companies still want clearer antitrust guidance on what can be shared. The next catalyst is whether Washington formalizes those rules before a major DeepSeek model upgrade forces another round of defensive measures.
