DeepSeek, Moonshot reportedly distilled GPT-6 Astra

DeepSeek and Moonshot AI relied heavily on outputs from OpenAI’s GPT-6 Astra to train their newest models, The Information reported on October 28, citing people with direct knowledge of the matter. The claim about DeepSeek rests on those unnamed sources alone. OpenAI’s own on-record attribution, from September 30, points only at individuals associated with Moonshot.
Two models are named in the report: Moonshot’s Kimi K3.1 Pro, released last week, and DeepSeek’s V4.1-Flash 1017, launched October 17. According to the report, both benefited from distilling Astra. Distillation means training one model on another model’s outputs. Anthropic, OpenAI and Google have all argued that it lets rivals reproduce advanced capabilities more quickly and at lower cost, and it has become a flashpoint as US-China technology competition heats up.
OpenAI’s September 30 post, “Disrupting a coordinated model-distillation campaign,” described a campaign that began in the first week of July, which OpenAI says it shut down on July 28. Activity peaked on July 24 and 25, when OpenAI logged 16,000 requests matching an extraction pattern from more than 4,000 users. OpenAI’s footnote says those figures count attempted extractions, not necessarily successful ones. OpenAI also said it was unclear whether all the operators were tied to a single actor, and attributed a core cluster to people associated with Moonshot, the developer of Kimi. The models named on Wednesday both shipped in October, after the shutdown date OpenAI gives.
The OpenAI findings came just weeks after Anthropic accused several Chinese AI developers, Moonshot and Alibaba among them, of secretly using Claude to help train their own systems, CNBC reported. OpenAI said it shared what it found with other AI developers through the Frontier Model Forum and through government information-sharing channels.
Earlier this month, Anthropic said new measures meant to stop rivals copying Claude’s capabilities had cut the data harvested in distillation attacks by about 90% to 99%.
Up next
Claude Enterprise gets team budgets and “off-topic” flags
Each Claude Enterprise org gets up to 10 free smart reports a month, per Mashable, and each one tags “personal or off-topic” sessions as inefficiencies. Would you keep personal chats in a work Claude account or move them elsewhere?
Related stories
- OpenAI ties a reasoning-extraction campaign to Moonshot AI
- DeepSeek, Kimi models carry Blackwell GPU fingerprint
- Distillation attack hit 1.4% of OpenCode Console users
- OpenAI flags Chinese models as risky after its own breaches
- OpenAI’s agents were poking at Hugging Face back in May
- OpenAI’s Brockman flags GLM-5.3 as a cyber risk
Comments
No comments yet. Be the first.
Join the conversation
Sign in to leave a comment. It appears after moderation.

