OpenAI: Moonshot AI users attempted to extract GPT reasoning data
OpenAI has accused users linked to China-based Moonshot AI of attempting to extract reasoning data from its GPT systems. The company stated the efforts aimed to replicate the capabilities of its advanced AI models and occurred in July.
Artificial intelligence firm OpenAI announced on September 30 that users associated with Moonshot AI, a Beijing-based startup behind the Kimi chatbot, allegedly attempted to extract hidden reasoning data from its GPT systems. OpenAI stated that these efforts could help reproduce the reasoning and capabilities of its most advanced models.
The alleged activity began in early July and peaked later that month, with OpenAI reporting up to 16,000 requests. According to the company, users copied encrypted reasoning from one conversation and asked a model in another conversation to transcribe it. This process could allow for the replication of OpenAI's proprietary AI functionalities.
OpenAI has shared details of the incident through the Frontier Model Forum, an industry group, and via government channels. This accusation adds to broader claims by the U.S. regarding Chinese AI developers potentially extracting proprietary knowledge from American firms.
The company did not publicly detail how any extracted material was ultimately used. The incident highlights the competitive landscape in AI development and the ongoing challenges of protecting advanced model data and intellectual property.