US government agencies have accused Chinese AI companies of secretly extracting billions of tokens from models built by OpenAI, Anthropic, Google, and SpaceX-owned Grok, according to Dark Reading. The alleged technique is model distillation, in which outputs from a large, advanced model are used as training data to build a cheaper, smaller model.
If the accusation holds, the practice would let a developer replicate much of a frontier model's behavior without paying for the expensive data collection and compute that originally produced it. The source frames the extraction as a way for Chinese firms to lower development costs and accelerate their own frontier-model efforts.
Because the article relies on a single US government accusation, it does not include independent confirmation or a response from the accused companies. The details remain allegations rather than established facts.