Chinese AI developers DeepSeek and Moonshot AI relied heavily on outputs from OpenAI’s GPT-6 Astra model to train their latest systems, according to a new report from The Information cited by Reuters. The practice, known as distillation, involves using a leading model’s answers to teach a smaller or rival system, and it has become one of the most sensitive flashpoints in the global AI race.
According to the report, Moonshot AI’s Kimi K3.1 Pro, released last week, and DeepSeek’s upgraded V4.1-Flash 1017 model, launched on October 17, both benefited from distilling Astra. Neither OpenAI, DeepSeek, nor Moonshot AI immediately responded to Reuters’ requests for comment, so the claims remain attributed to unnamed people with direct knowledge rather than confirmed by the companies themselves.
The dispute matters because distillation can let a challenger reproduce advanced capabilities far more quickly and cheaply than building them from scratch. Leading labs argue that this undermines the enormous investment behind frontier models such as ChatGPT and Claude, while rivals say the technique is a standard way to improve smaller, more efficient systems like DeepSeek.
OpenAI had already signaled its concern. In a September 30 blog post, the company said it had linked a core part of a distillation campaign that began in July to individuals associated with Moonshot AI. Earlier this month, Anthropic said new defenses against distillation attacks had cut harvested data from its Claude models by about 90% to 99%.
As US-China technology competition intensifies, the latest report is likely to sharpen calls for stronger technical safeguards and clearer rules on what rival labs may learn from each other’s models.
Source: Reuters via MIX-FM
