DeepSeek and Moonshot Used OpenAI GPT-6 Astra to Train Latest Models, The Information Reports

Report points to large-scale use of model outputs

Chinese AI developers DeepSeek and Moonshot AI relied heavily on outputs from OpenAI's GPT-6 Astra model to train their latest models, The Information reported on Wednesday, citing people with direct knowledge of the matter.

The report did not establish the full scope of the alleged use or identify the specific models involved. DeepSeek, Moonshot AI and OpenAI did not immediately respond to Reuters' requests for comment.

Distillation dispute intensifies

The practice described in the report is commonly known as distillation: using one model's outputs to help train, reproduce or improve another model. AI companies have increasingly treated the practice as a flashpoint because it can let competitors reproduce advanced capabilities more quickly and at lower cost.

OpenAI said in a September 30 post that it had identified and disrupted a coordinated campaign to extract protected reasoning from its models. The company said the activity began in early July, included a surge of 16,000 requests from more than 4,000 users, and was linked in part to individuals associated with Moonshot AI, the developer of Kimi. OpenAI also said it was unclear whether every operator belonged to one actor.

OpenAI said the campaign involved manipulating model interactions rather than breaching its encryption, databases or stored user conversations. It described the activity as adversarial distillation and said it had shared relevant findings with industry and government partners.

Wider technology rivalry

The allegations add to tensions in the AI industry as U.S. and Chinese companies compete over increasingly capable systems. They also highlight the unresolved boundary between legitimate evaluation or research and the unauthorized extraction of model capabilities.

Sources

Reuters report
OpenAI: Disrupting a coordinated model-distillation campaign
Moonshot AI: About