NYU Professor Warns Against Sharing Ideas with OpenAI After Research Leak

A New York University mathematician has claimed that OpenAI's model reached his unpublished research findings within one day of learning his name, leading him to announce he will no longer input his original ideas into the company's AI tools. OpenAI clarified that his prompt could not have influenced their system.
Tristan Buckmaster, a mathematician at NYU, issued this warning several days after OpenAI published 722 AI-written math papers. He revealed one of his accounts was still set to enable model training by default.
Buckmaster and mathematician Levent Alpoglu from competitor Anthropic spent nearly a year working on a step toward solving the million-dollar fluid dynamics problem, Navier-Stokes equations. Buckmaster used OpenAI's coding assistant Codex through a paid personal account to run this research, with drafts stored on OpenAI servers.
He told CNBC that no progress was made before OpenAI learned his name, but within one day of that, OpenAI's agent reached 'exactly where we were.' Buckmaster stated in September that he wasn't blaming anyone, per BeInCrypto, though he now finds the timing suspicious.
Before OpenAI announced its Navier-Stokes proof on September 8, Buckmaster said the company proposed listing him as sole author excluding Alpoglu. He declined the offer. OpenAI denied any connection to this data.
Buckmaster noted the timeline kept changing; later company statements fixed it at July 3.
He advises researchers not to input ideas into AI tools, especially for creative work he wants to keep original. He highlighted that grant reviewers often paste research plans into AI tools, risking leaks of unpublished work to AI companies.
OpenAI's October release cited 4,000 unresolved problems as the source. Buckmaster claims this has caused many junior mathematicians' programs to disappear. OpenAI Senior Researcher Sébastien Bubeck told The New York Times that these proposals were made with good intentions to help Buckmaster.
An OpenAI spokesperson stated the company had made 'significant progress' on another million-dollar problem.
Korean Source
This article is an English localization of a Korean-language crypto news report. Original headline: 챗GPT, 아이디어 맡겨도 될까…NYU 교수 ‘신중해야’