OpenAI Says It Halted a Reasoning-Extraction Campaign, Citing Moonshot AI Connections
Decrypt reports OpenAI says it stopped a campaign that pulled hidden reasoning out of its models, tracing a core cluster to people linked to Moonshot AI.
Why it matters
The report touches no crypto asset, so its significance for digital-asset markets is limited. It matters to AI labs because OpenAI describes a method that needed neither broken encryption nor database access, and says the extracted reasoning could train a model without the original safeguards.
Key facts
OpenAI dates the start of the extraction campaign to July 1.
Reported by Decrypt
OpenAI says it logged 16,000 extraction requests on July 24 and 25.
16,000 requestsReported by Decrypt
Those requests on July 24 and 25 came from more than 4,000 users.
4,000 usersReported by Decrypt
The requests formed part of a wider cluster of more than 15,000 users.
15,000 usersReported by Decrypt
OpenAI attributes a core cluster of the activity to individuals associated with Moonshot AI.
Reported by Decrypt
OpenAI says it fully disrupted the activity by July 28.
Reported by Decrypt
OpenAI says the operators did not break its encryption, compromise a database or gain direct access to stored user conversations, and manipulated model interactions instead.
Reported by Decrypt
OpenAI says the extracted hidden reasoning could be used to train another model without the original safeguards.
Reported by Decrypt
What happened
Decrypt reported that OpenAI says it shut down a coordinated campaign aimed at the hidden reasoning its models generate before answering, and that it traces a core cluster of that activity to individuals associated with Moonshot AI, the Chinese startup behind the Kimi chatbot. OpenAI dates the start of the campaign to July 1 and says it logged 16,000 extraction requests from more than 4,000 users on July 24 and 25, part of a wider cluster of over 15,000 users. Decrypt reported that OpenAI says it fully disrupted the activity by July 28. The company said the operators did not break its encryption, compromise a database or reach stored user conversations, and instead manipulated model interactions so that protected reasoning came out, which OpenAI says could be used to train another model without the original safeguards. OpenAI also says it is unclear whether every operator in the cluster came from a single actor.
How the story developed
- First report
First report by Decrypt
OpenAI Says People Linked to China's Moonshot Tried to Copy Its AI's Hidden Reasoning
- Update
Decrypt reports OpenAI's account of a reasoning-extraction campaign
Single-source report describing what OpenAI says it detected and stopped.
- Official statement
OpenAI dates the start of the campaign to July 1
No year is given in the report.
- Resolution
OpenAI says it fully disrupted the activity by July 28
The company states the operation was shut down.
- Update
Peak activity on July 24 and 25
16,000 extraction requests from more than 4,000 users, within a cluster of more than 15,000 users.
- Official statement
OpenAI attributes a core cluster to Moonshot AI-linked individuals
OpenAI also says it is unclear whether all operators came from a single actor.
Affected entities
Original sources
| Publisher | Report | Role | Published |
|---|---|---|---|
| Decrypt Crypto media | OpenAI Says People Linked to China's Moonshot Tried to Copy Its AI's Hidden Reasoning | Original |