Disrupting a coordinated model-distillation campaign
OpenAI says it disrupted a campaign extracting protected reasoning from its models and attributes a core cluster to people associated with Moonshot AI.
Activity from 2026-07-01 spiked on 2026-07-24 and 07-25 with 16,000 requests from over 4,000 users; related patterns reached over 15,000 users, disrupted by 07-28. Attackers replayed encrypted reasoning across conversations. OpenAI hardened hidden-reasoning protections.
- Date
- Wednesday, 30 September 2026
- Lab
- OpenAI
- Kind
- paper
- Access
- research preview
Figures
| Measure | Value | Measured by |
|---|---|---|
| Users in the related prompt-pattern cluster | more than 15,000 disrupted by 2026-07-28 | company |
Attribution to Moonshot AI is OpenAI's own assessment ("a core cluster"); the post says it is unclear whether all operators were one actor, and Moonshot's response was not found. Distillation as a diffusion route for reasoning capability is the core issue for B20.
Sources
- openai.com/index/disrupting-a-coordinated-model-distillation-campai/
- thehackernews.com/2026/10/openai-disrupts-reasoning-extraction.html
- decrypt.co/379890/openai-china-moonshot-copy-ai-hidden-reasoning
This record was checked against its sources on 6 October 2026. How we check