Reuters reported on October 29, 2026, citing The Information, an allegation that DeepSeek and Moonshot AI had used responses from OpenAI’s GPT-6 Astra to train newer models. The claim concerns generated responses, not access to GPT-6 Astra’s model weights. Techzine reported that OpenAI, DeepSeek, and Moonshot AI had not immediately responded to Reuters’ requests for comment as of October 29.
The two model versions named
Techzine identified DeepSeek V4.1-Flash 1017 and Kimi K3.1 Pro, from Moonshot AI, as the models named in coverage of the allegation.
What distillation means here
Distillation is a training approach in which one model learns from another model’s outputs. In this case, the reported claim concerns GPT-6 Astra’s responses as training material, not its weights.
An earlier OpenAI campaign was a separate episode
Techzine reported that an OpenAI post dated September 30, 2026, linked a component of a separate campaign, said to have begun in July, to individuals associated with Moonshot AI. The account also described OpenAI restricting accounts, tightening registration checks, and expanding network monitoring. NeoTeo previously covered separate allegations involving Claude and Chinese AI firms.
Anthropic says Claude was used in biology, influence and cyber operations
