Second Mathematician Accuses OpenAI of Opacity Over Training Data for Math Breakthroughs

A growing dispute over the provenance of OpenAI‘s mathematical achievements has drawn in a second researcher, intensifying scrutiny of how the company sources the data behind its models’ increasingly notable results in advanced mathematics.

Days after an earlier confrontation over whether OpenAI’s systems drew on unpublished academic work, mathematician Andreas Thom publicly accused the company of “dishonest” conduct and insufficient transparency. In a series of posts on Mastodon, Thom raised the possibility that exchanges he and his colleagues had with ChatGPT prior to OpenAI’s announcement of its mathematical results may have fed into the models’ performance in the field.

The controversy centers on a question that has dogged the AI industry more broadly: when a model produces a genuine research-level result, how much of the underlying knowledge came from public publications, and how much came from private interactions, preprints, or work shared in confidence? For mathematicians — a community that often circulates unpublished proofs and drafts among trusted colleagues — the stakes are reputational as much as commercial.

Neither accusation has been independently verified, and OpenAI has not publicly detailed the specific composition of the datasets used to train its frontier models, a practice consistent with the company’s general approach to disclosing training data. That lack of visibility is precisely what the researchers say they are seeking to change, calling for proof that their work was not used without attribution or consent.

The episode lands amid heightened sensitivity across the technology sector over data provenance and intellectual property. It also arrives at a moment when public-market investors are watching the AI ecosystem closely. Among tech names moving today, semiconductor component maker Arris-listed ARRYY — Array Technologies (ARRY) — traded at $4.48, up 0.45% from its previous close of $4.46, with a market capitalization of roughly $689 million. Meanwhile, payments and fintech firm Block, Inc. (XYZ) slipped 1.0% to $79.40 from a prior close of $80.20, valuing the Square and Cash App parent at approximately $47.7 billion. Neither company is directly involved in the dispute, but both illustrate the breadth of the public technology market now shaped by debates over AI data practices and platform trust.

For OpenAI, the math community’s complaints represent a narrow but vocal constituency with outsized influence over how the company’s research claims are perceived. Resolving the question of training-data provenance may matter less for current revenue than for the credibility of future claims about model-driven scientific discovery — claims that are becoming a central part of the company’s public narrative.

What to watch

• Whether OpenAI issues a formal response or publishes details about its mathematics training data sources.
• Any additional researchers coming forward with similar claims about private interactions with ChatGPT.
• Upcoming OpenAI research announcements and whether attribution or data-provenance practices change.
• Broader industry or regulatory movement on AI training-data disclosure requirements.

Source: original release.