OpenAI Says Moonshot-Linked Accounts Tried to Extract Hidden Model Reasoning

OpenAI says operators linked to China's Moonshot AI attempted to extract hidden reasoning from its models.
According to OpenAI, the operators copied encrypted reasoning from one conversation and asked another model to decrypt it.
At its peak in late July, the campaign generated roughly 16,000 requests from more than 4,000 accounts before OpenAI shut it down.
The disclosure comes after Anthropic previously accused Moonshot of routing user requests through Claude to help train its own models. Moonshot has not yet responded to the latest allegation.