OpenAI, Anthropic, Google Form Alliance to Combat Chinese AI Model Distillation and Cloning

OpenAI, Anthropic, and Google announced cooperation through the Frontier Model Forum on April 7 to detect and prevent adversarial distillation attempts and model extraction violations by Chinese AI labs. The companies have observed systematic use of frontier Western models to generate training data for Chinese models — a technique known as knowledge distillation — which they argue violates terms of service and undermines US export control intent. The joint initiative includes: shared technical signatures to detect when queries are designed for systematic extraction rather than legitimate use; coordinated enforcement against high-volume API abuse; and a protocol for flagging suspected state-directed distillation campaigns to relevant government agencies. The announcement followed public accusations that DeepSeek's V3 model was trained partly using OpenAI API outputs, which DeepSeek denied — but which OpenAI stated it was investigating. The alliance reflects a view among US AI labs that technical excellence alone is insufficient to maintain competitive advantage when models can be systematically extracted at low cost. Critics noted the irony that one of the main arguments for AI openness — that knowledge is inherently non-rival — is being used to justify restrictions on what knowledge the frontier models can share.