Artificial Intelligence
OpenAI accuses Moonshot AI over model distillation
OpenAI said it linked a July campaign to people associated with Moonshot AI, described the operators as trying to copy hidden model reasoning, and said it later blocked the activity and tightened safeguards.
- OpenAI said people linked to Moonshot AI were behind a campaign to extract the hidden reasoning of its models.
- It said the activity began on 1 July, surged on 24 and 25 July, and was shut down by 28 July.
- OpenAI said the operators copied encrypted reasoning from one conversation and used another model to decrypt and transcribe it.
- OpenAI said it did not find evidence of broken encryption, database access or exposure of stored user conversations, and it later banned accounts and tightened sign-up checks.
OpenAI said it had found a coordinated campaign linked to people associated with Moonshot AI, the Chinese developer of Kimi. It said not every operator could be tied to one actor, but it attributed a core cluster to people associated with Moonshot.
The company said the activity began on 1 July and rose sharply on 24 and 25 July, when it saw 16,000 requests from more than 4,000 users. OpenAI said it later found related activity across more than 15,000 users and shut the campaign down by 28 July.
OpenAI said the operators copied encrypted reasoning from one chat and then used a separate model to decode and write it out. The company said they did not break encryption, access a database or reach stored conversations.
OpenAI said it had banned accounts, tightened sign-up checks and added protections around hidden reasoning. Caroline Zier, who leads strategic national security policy initiatives at OpenAI, told Bloomberg that the concern was terms-of-service violations rather than open models or legitimate distillation.
The article said Bloomberg reported this was the first time OpenAI had accused Moonshot. It also said Anthropic had made a similar claim, US agencies had issued an advisory with the same allegation, China had rejected that advisory, and David Sacks had said such reporting was meant to push the US towards banning rival open models.
Named in this story
People
- Caroline Zier
- OpenAI policy lead who said the concern was terms-of-service breaches
- David Sacks
- Trump’s former AI czar who criticised the reports
Companies
- OpenAI
- said people linked to Moonshot AI ran the campaign and published its findings
- Moonshot AI
- Chinese developer of Kimi that OpenAI linked to the activity
- Anthropic
- had made a similar claim against Moonshot AI
Organisations
- Frontier Model Forum
- channel through which OpenAI shared its findings with other AI labs
Governments and agencies
- US agencies
- issued an advisory making the same charge
- China
- rejected the advisory as unfounded
Products and systems
- Kimi
- Moonshot AI’s model family named in the article
How the source tells it
Mostly straight reporting, with some conflict framing and security-language emphasis supplied by attributed sources.
- threat OpenAI’s own terminology cast the activity as extraction of hidden model reasoning
- conflict the article was organised around accusation, rebuttal and outside scrutiny