OpenAI Accuses Moonshot AI-Linked Users of Distillation Campaign Targeting Hidden Reasoning
The company says thousands of accounts attempted to extract encrypted chain-of-thought outputs before being shut down in late July.
OpenAI announced on Wednesday that individuals associated with Chinese developer Moonshot AI conducted a coordinated campaign to extract internal reasoning logs from its models, according to a report by The Next Web (https://thenextweb.com/news/openai-moonshot-distillation-campaign-hidden-reasoning).
The activity started at low volume on July 1 and peaked on July 24 and 25, generating 16,000 requests across more than 4,000 user accounts. OpenAI identified related activity across more than 15,000 users before shutting the campaign down on July 28. While OpenAI stated it could not link every operator to a single actor, it attributed a core cluster of the accounts to people linked to Moonshot, the developer of the Kimi model.
OpenAI characterized the operation as adversarial distillation—the unauthorized use of one model's outputs or reasoning chains to train, reproduce, or enhance another model. OpenAI models generate encrypted reasoning chains prior to answering; operators copied this encrypted text from one conversation and prompted the model in a separate session to decrypt and transcribe the underlying reasoning.
The company stated that the operators did not break encryption, breach databases, or access stored user conversations. In response, OpenAI banned the involved accounts, added technical safeguards around hidden reasoning, tightened sign-up verification, and shared intelligence with peer companies via the Frontier Model Forum and with government channels.
"Our concern is about violation of our terms of service, not open models or legitimate distillation," Caroline Zier, lead for strategic national security policy initiatives at OpenAI, told Bloomberg.
The attribution marks the first time OpenAI has publicly accused Moonshot, following similar claims by Anthropic last month and a recent U.S. government advisory naming six Chinese firms. China rejected the U.S. advisory as unfounded, while former Trump AI official David Sacks told Bloomberg that such reports represent efforts to lobby for bans against foreign open models. Moonshot and peer DeepSeek are also currently under investigation by Chinese domestic regulators, according to The Information.
Sources
Written by
The Company Wire
Inside the companies building what’s next. Reporting on startups, technology, funding and the people shaping them.
