Report Claims DeepSeek and Moonshot AI Used OpenAI’s GPT-6 Astra Outputs to Train New Models

Reported claims about model distillation

A report by The Information, citing people with direct knowledge of the matter, has claimed that two Chinese AI developers—DeepSeek and Moonshot AI—relied on outputs from OpenAI’s GPT-6 Astra to build their latest systems. Specifically, the report says Moonshot AI’s Kimi K3.1 Pro and DeepSeek’s V4.1-Flash 1017, launched on October 17, drew on distilled data derived from GPT-6 Astra. OpenAI described GPT-6 Astra in September as its “world’s most intelligent and aligned model.”

These are reported claims, not independently confirmed findings in this article.

OpenAI’s earlier warning

OpenAI said in a September 30 blog post that it had traced a core part of a broad distillation operation, running since July, to individuals affiliated with Moonshot AI. The source material does not include the full evidence behind that tracing.

Industry context

Model distillation refers to training a smaller or newer system on the outputs of a larger model. American tech companies have argued that systematically extracting synthetic responses lets competitors replicate advanced capabilities faster and at lower cost than training foundation models from scratch. Frontier AI labs including OpenAI, Anthropic, and Alphabet have warned about unauthorized distillation.

Anthropic disclosed earlier this month that it deployed defensive countermeasures against attempts to replicate capabilities from its Claude model family. According to Anthropic, those protections reduced data harvested through distillation attempts by approximately 90% to 99%.

Practical implications

• If distillation claims are substantiated, it could intensify US-China AI competition and legal scrutiny over training data.
• Developers may face pressure to strengthen technical guardrails against automated extraction.
• Users and enterprises should treat performance claims about new models with caution until independent benchmarks or disclosures are available.

How to verify

• Read The Information’s original report for sourcing details.
• Review OpenAI’s September 30 blog post on unauthorized scraping.
• Check Anthropic’s public disclosure on Claude countermeasures.
• Look for statements from DeepSeek and Moonshot AI responding to the allegations.

This is based on published reports; this article does not claim firsthand reporting or independent verification of the distillation allegations.