OpenAI’s Astra used by Chinese AI firms to train models- The Information

Chinese AI companies DeepSeek and Moonshot AI have been using outputs from OpenAI's advanced GPT-6 Astra model to train their newest AI models, according to a report by The Information. Distilling, as this practice is known, has become a contentious issue as the US-China tech rivalry heats up. Both companies claim it enables them to replicate sophisticated AI capabilities more swiftly and at a reduced cost.

Moonshot AI's Kimi K3.1 Pro model was released recently, while DeepSeek's upgraded V4.1-Flash 1017 model was launched on October 17. OpenAI, DeepSeek, and Moonshot AI have yet to comment on the matter. In September, OpenAI acknowledged a link between a distillation campaign that started in July and individuals linked to Moonshot AI.

Major tech firms such as Anthropic, OpenAI, and Google also endorse distillation, stating it allows competitors to duplicate advanced AI features more rapidly and cost-effectively. Recently, Anthropic announced new measures aimed at preventing rivals from copying the capabilities of its Claude AI models, which have reportedly reduced data harvested in distillation attacks by 90% to 99%.