A second mathematician has publicly accused OpenAI of “dishonest” behavior and a lack of transparency over whether user interactions contributed to the company’s recent mathematical breakthroughs, adding to growing unease in academic circles in September 2026.
Andreas Thom, a mathematician, raised concerns in a series of Mastodon posts that conversations he and colleagues had with ChatGPT may have fed into OpenAI’s training data — and ultimately helped the company achieve results in his own area of expertise. One of the ten mathematical results OpenAI announced last month involved non-sofic groups, a field in which Thom works alongside mathematician Gábor Kun. OpenAI had initially failed to acknowledge their contributions before quietly amending its writeup after criticism from the mathematical community.
Thom said he was struck by “OpenAI’s detailed command of our techniques,” which he described as neither the most obvious nor the most promising approaches at the time. He emailed OpenAI researchers Sébastien Bubeck and Mark Sellke — also a statistician at Harvard — to ask whether his ChatGPT conversations had entered the company’s training pipeline. He said the response only addressed whether his conversations could be accessed directly, not whether they had influenced model training. “I take this as dishonesty to say the least,” Thom wrote.
His concerns follow those of Tristan Buckmaster, a mathematics professor at New York University, who had publicly questioned whether his use of OpenAI’s Codex tool influenced the company’s models. In its announcement of a solution to the Navier-Stokes Millennium Prize problem, OpenAI denied accessing specific user data but acknowledged it “cannot rule out that de-identified data derived from their usage of our products helped improve our models.” Thom called this distinction misleading, noting that “de-identification may remove a name; it does not remove the intellectual content of a mathematical idea.”
Thom said it “would be ethically indefensible” if nonpublic research shared by users helped improve models that OpenAI then used to race those same researchers to publication without consent, disclosure, or credit. He argued that only OpenAI holds the data needed to resolve the question, and that the burden of proof lies with the company.
Multiple researchers told The Verge they worry the pattern of behavior could push mathematics into greater secrecy, as academics may become reluctant to share progress if even rumors of a breakthrough could prompt a well-resourced competitor to pursue the same problem. OpenAI did not respond to a request for comment.
Source: The Verge