OpenAI said on September 30 that users it linked to Moonshot AI, a Beijing-based AI startup behind the Kimi chatbot, tried to extract hidden reasoning data from its GPT systems.

OpenAI said the effort could help reproduce the reasoning and capabilities of its most advanced models.

OpenAI said the activity began in early July and peaked later that month at 16,000 requests. The company said users associated with Moonshot copied encrypted reasoning from one conversation and asked a model in another conversation to transcribe it.

OpenAI said it shared details of the incident through the Frontier Model Forum, an industry group, and government channels.

The allegation adds to broader claims by the US that Chinese AI developers are extracting proprietary knowledge from American firms. OpenAI did not publicly detail how any extracted material was ultimately used.

🔗 Source: Bloomberg

Recent OpenAI developments