OpenAI’s GPT-6 Astra
OpenAI released GPT-6 Astra in September as the ‘world’s most intelligent and aligned model’ and two months after the launch, a report has claimed that two prominent Chinese artificial intelligence (AI) developers, DeepSeek and Moonshot AI, relied heavily on outputs from the company’s models to build their latest systems.
Citing people with direct knowledge of the matter, The Information reported that both Moonshot AI’s newly unveiled Kimi K3.1 Pro and DeepSeek’s updated V4.1-Flash 1017 model, launched on October 17, drew on distilled data derived from GPT-6 Astra. In a blog post published on September 30, the San Francisco-based lab stated that it had directly traced a core section of a broad distillation operation, running since July, to individuals affiliated with Moonshot AI. Those updated security protections have cut down data harvested through distillation attempts by approximately 90% to 99%, according to Anthropic.
The technique, widely referred to across the industry as model distillation, has turned into a major flashpoint between global tech labs as competition between the US and China intensifies. Frontier lab leaders, including OpenAI, Anthropic, and Google parent Alphabet, have frequently warned that unauthorised distillation enables competitor firms to capture advanced capabilities without incurring the steep costs required to train foundation models from scratch.
OpenAI had already sounded the alarm on unauthorised model scraping earlier this year. Earlier this month, Anthropic disclosed that it had deployed defensive countermeasures specifically designed to prevent rival companies from replicating capabilities from its Claude model family.