Artificial intelligence laboratories are pushing deeper into advanced mathematics, but the speed of their progress is increasingly inseparable from a dispute over how that progress is presented. In a report updated October 5, The Verge chronicled a year in which OpenAI, Anthropic, and other labs announced results on long-standing problems, including claims that exceeded what many researchers thought current systems could accomplish. The technical stakes are high, yet much of the argument now centers on process: who receives credit, what evidence is disclosed, and whether an industrial race can coexist with academic norms.

The most dramatic example is OpenAI’s claim that an internal model found a solution to the Navier–Stokes problem, one of the seven Millennium Prize Problems. According to The Verge, OpenAI said the work used a model more capable than its newly released GPT-6 Astra together with 10,000 concurrent agents. The company said training for that internal system began on August 28. The claimed result would be extraordinary if it survives the field’s scrutiny; the Clay Mathematics Institute attaches a $1 million prize to each Millennium problem, but formal recognition requires far more than a company announcement.

Two streams of equations converge on one unfinished proof beneath a stopwatch shadow.
The dispute extends beyond whether a proof is correct to questions about priority, process, and how competing work is treated.

The announcement did not arrive in a vacuum. The Verge reported that OpenAI accelerated its effort after learning that other researchers were making progress, prompting allegations within the mathematics community about scooping, surveillance, and breaches of professional custom. Those accusations do not by themselves settle the validity of the proof. They do show that a mathematically correct answer would not resolve the wider conflict over how the answer was obtained and released.

A separate dispute concerns the provenance of the knowledge behind these systems. Mathematician Andreas Thom questioned whether conversations that he and colleagues had with ChatGPT may have aided OpenAI’s work, The Verge reported. One of ten advances publicized by the company involved non-sofic groups and drew heavily, by OpenAI’s own acknowledgment, on earlier work by Thom and Gábor Kun. The episode sharpened demands for greater transparency about training data and about the boundary between learning from published research and benefiting from researchers’ unpublished ideas.

OpenAI has responded to the strained relationship by announcing an independent advisory group of prominent mathematicians. The panel is intended to advise the company and other AI developers on interactions with mathematical research, including the presentation and release of new results. Researchers interviewed by The Verge described the initiative as a potentially useful first step, while questioning its remit, its independence in practice, and whether a small group can speak for a broad global discipline.

A large stack of AI-generated proofs approaches a narrow peer-review gateway.
A surge of machine-assisted results could strain the human systems used to verify proofs and establish reliable credit.

The advisory group may face an immediate stress test. The Verge reported that it was expected to help coordinate the release of scores of additional results produced by an unreleased OpenAI model. That prospect has generated anxiety among researchers who worry that a rapid sequence of machine-assisted claims could overwhelm the slower systems used to check proofs, establish priority, and assign credit. Verification remains essential because a persuasive-looking derivation is not the same thing as a proof accepted by specialists.

For mathematicians, the tension is not simply a contest between people and machines. AI systems can connect methods across subfields, surface obscure literature, and propose routes through difficult problems, according to the account. Those capabilities could widen exploration and shorten the path to discovery. But if labs treat mathematics chiefly as a benchmark or a race for first place, researchers fear the incentives will favor dramatic announcements over careful collaboration and durable confidence in the results.

The next phase will therefore be judged on two tracks. OpenAI and its peers must show that their claimed advances withstand expert review, and they must also demonstrate that their methods of disclosure, attribution, and engagement deserve trust. The Verge’s reporting suggests the industry recognizes that its earlier approach damaged relationships. Whether an advisory panel and promises of better coordination can repair them remains uncertain, especially as more results are waiting to be released.