Chinese AI giants DeepSeek, Moonshot used OpenAI's GPT-6 Astra to train their models: Report

The claims renew concerns about AI distillation, which can help developers replicate advanced model capabilities more quickly and cheaply.

Chinese AI firms DeepSeek and Moonshot AI have reportedly used outputs from OpenAI’s advanced GPT-6 Astra model to train their latest AI systems, according to a report by The Information on Wednesday.

In a blog post published on September 30, OpenAI said it had traced a significant part of a distillation campaign that began in July to individuals linked to Moonshot AI.

DeepSeek and Moonshot AI reportedly made extensive use of Astra’s outputs to train their latest models, sources familiar with the matter told The Information.

AI distillation remains a major reason for progress in Chinese open-weight models, despite repeated criticism of the practice from OpenAI and Anthropic.

The Trump administration had barred civilian US government agencies from using products and services offered by Chinese AI startup DeepSeek, Bloomberg reported earlier. The decision was driven by concerns over potential national security risks associated with the company’s operations in China.

What is AI distillation?

The use of AI distillation has sparked growing concerns among major technology companies as competition between the United States and China intensifies. The technique allows developers to learn from the outputs of advanced AI models, potentially enabling them to build competing systems faster and at a lower cost.

According to OpenAI, adversarial distillation is “the systematic and unauthorized use of one model’s outputs or reasoning to help train, reproduce, or improve another model.”

Major AI developers, including Anthropic, OpenAI and Google, have warned that the practice could allow competitors to replicate the capabilities of their advanced models without investing similar resources.

Earlier this month, Anthropic reportedly announced that new safeguards intended to prevent the unauthorised replication of its Claude AI models had cut the volume of data gathered through suspected distillation attacks by around 90% to 99%.

OpenAI traced distillation to Moonshot AI

In a blog post published on September 30, OpenAI said it had traced a significant part of a distillation campaign that began in July to individuals linked to Moonshot AI, the developer of Kimi.

It alleged that the operators “manipulated model interactions so that protected reasoning could be reproduced in forms visible to the requester in a coordinated, scaled manner that violated our terms of service.”

OpenAI said the activity began on July 1, with a sharp rise to 16,000 requests on July 24 and 25 involving more than 4,000 users. Further investigation uncovered related activity across a network of over 15,000 users, which OpenAI said it had fully disrupted by July 28.

(with inputs from Reuters, Bloomberg)