OpenAI says it flagged Chinese-linked effort to extract AI model reasoning
OpenAI says it identified an attempt to extract protected reasoning processes from its AI models, with part of the activity linked to China's Moonshot AI.
Written by EGazette’s AI. The facts are drawn from cited sources; the analysis is the AI’s own.
OpenAI has said it identified an attempt to extract protected reasoning data from its artificial intelligence models, linking part of the activity to Moonshot AI, a Chinese AI company known for its Kimi model line.
According to OpenAI, the effort targeted the underlying reasoning processes that power its models' outputs, rather than simply attempting to replicate the models' visible responses. Reasoning data of this kind is generally considered commercially sensitive by AI developers, since it underpins how a model arrives at its answers.
OpenAI did not provide extensive technical detail on how the extraction attempt was carried out or how it was detected, but the company said it had identified the activity and linked a portion of it to the Chinese firm.
The disclosure comes amid broader industry scrutiny of competition between US and Chinese AI developers, with companies on both sides racing to build more capable models while also working to protect proprietary techniques from being copied or reverse-engineered by rivals.
Moonshot AI has not been quoted in connection with OpenAI's claim in the reporting of this story, and it remains unclear whether the company has responded to the allegation.
Sources
- OpenAI flags Chinese-linked effort to extract AI model reasoning — CNBC — Top News
EGazette summarizes reporting from multiple sources; follow the links for the originals.
Comments
Loading comments…
Related articles
Sam Altman Declines Senate Invitation on AI Risk Hearing
OpenAI's chief executive turned down a congressional request to testify about the dangers of advanced AI systems, according to the senator who extended the invitation.
Hawley says OpenAI's Sam Altman declined to testify at Senate hearing on 'rogue AI'
A Senate subcommittee examined the risks of "rogue AI" this week following a string of AI-related cyberattacks and mounting warnings about the technology.
Elon Musk's Grokipedia rolls out refreshed design in v0.3 update
Grokipedia, SpaceXAI's AI-powered Wikipedia competitor, has updated its design with a new logo and refreshed homepage as it resumes incorporating edits.
Experts Tell US Lawmakers AI Models Are Reasoning in Ways Beyond Human Oversight
Apollo Research CEO Marius Hobbhahn said an OpenAI model studied last year used language not fully understandable to humans.
China Cracks Down on AI Companion Chatbots, Raising Questions on Global Approach
Beijing's restrictions on AI systems that replicate human relationships have prompted debate over whether China is ahead on AI regulation.
Trump's new AI chatbot 'America' shifts its answers over time, review finds
The chatbot, launched as a digital portal for government information, has reportedly changed how it responds to questions that initially challenged some of the president's claims.