OpenAI says its AI agents solved one of mathematics' most stubborn open problems. Within days of the announcement, a growing number of mathematicians said the company may have solved it using their own unpublished work — and that nobody asked them first.

The dispute has escalated from a technical argument into an unusually bitter public fight over credit, training data provenance, and whether AI companies can claim intellectual breakthroughs built on the uncredited labor of the researchers they are ostensibly surpassing.

What OpenAI claims

Announced on a call with reporters rather than in a journal or at an academic conference, OpenAI's result was framed as a landmark: according to coverage aggregated by TechSpot, roughly 10,000 of the company's AI agents worked in parallel for about 88 hours to crack a problem that had resisted human mathematicians for nine decades. MSN reported the effort consumed 130 billion tokens. Decrypt and The Hindu described it as a million-dollar problem, though the precise identity and prize status of the question varies across reports — a reflection of how quickly the announcement outran the documentation behind it.

Unlike a conventional mathematical paper, there is no peer-reviewed manuscript to inspect. There is a press briefing, a set of company claims, and — as The Next Web noted pointedly — a contrast in verification cultures.

‘Unethical’ and ‘dishonest’

The backlash came fast. The Verge reported that mathematician Andreas Thom raised concerns in a series of Mastodon posts that interactions he and his colleagues had had with ChatGPT before OpenAI's triumphant announcement may have contributed directly to its success.

A second mathematician went further. Coverage from the New York Post and MSN describes a professor — characterized as an NYU mathematician — claiming the company cribbed his work and then threatened him when he objected. Yahoo's aggregation of the story put it bluntly: experts say OpenAI's latest math breakthroughs commit research misconduct.

The Verge reported that a mathematician accused the AI giant of unethical and “dishonest” behavior and a lack of transparency about the origins of its training data.

That last charge is the most structurally serious. It is not simply a question of whether OpenAI used someone's theorem without citing it. It is a question of whether unpublished, non-public mathematical reasoning — volunteered in conversations with a chatbot — became training signal for a system that then beat its authors to the finish line.

Machine-checkable proofs vs. a press call

The Next Web distilled the asymmetry into a single line: the mathematicians published machine-checkable proofs; OpenAI announced its result on a call with reporters.

In mathematics, a machine-checkable proof — formalized in software like Lean or Coq — is close to the gold standard of verification. It can be audited line by line by anyone. A press call cannot. Whether OpenAI's agents produced an equivalent artifact, and whether it has been released for independent inspection, remains among the central unanswered questions.

How different outlets frame the story

  • The Verge frames it as a second front in an ongoing ethics war, noting this is the second mathematician in days to accuse OpenAI of benefiting from unpublished work, and describing the episode as sending “a chill through academia.”
  • Wired leads with institutional skepticism: a huge math discovery, and academics crying foul.
  • MIT Technology Review zooms out, asking what the controversy reveals about the future of mathematics as a discipline.
  • The Detroit News offers the starkest framing of all: this may be the first academic profession to see its work taken over by AI.
  • Crypto and business outlets (Decrypt, BeInCrypto) foreground the money — the million-dollar prize framing — and the competitive race between OpenAI and rival researchers.

Why mathematicians are watching so closely

Mathematics has long operated on norms that predate AI: attribution, priority, and peer review. A result belongs to whoever proves it first and publicly, and the proof must survive scrutiny by others. OpenAI's approach inverts that sequence — announce first, verify later, and let the compute footprint stand in for the argument.

There is a deeper concern too. Unlike code or prose, where AI companies can point to public repositories and published corpora, frontier mathematical reasoning often lives in unpublished drafts, private emails, seminar conversations, and now chatbot sessions. If those channels feed model training without consent or compensation, the profession's informal knowledge commons becomes an extraction site.

The unanswered questions

  • What exactly was the problem, and where is the formal, machine-checkable proof?
  • Was any of the disputed material part of OpenAI's training corpus, and will the company disclose that?
  • What did the alleged threat to a professor consist of, and who made it?
  • Does a large-scale agentic run count as a mathematical contribution, or as a computation whose intellectual scaffolding was borrowed?

OpenAI has not resolved these questions publicly. Until it does, the episode is likely to become a reference point — the moment the AI industry's habit of announcing first and explaining later collided with a field that has spent 2,500 years insisting on the opposite.