Mathematicians Hate AI. They Can’t Quit It

The academic world, particularly the realm of mathematics, is grappling with a profound ethical and practical dilemma as artificial intelligence rapidly advances. At the heart of this storm is Tristan Buckmaster, a mathematician who alleges that tech giant OpenAI leveraged his ongoing research to prematurely claim a breakthrough in solving the Navier-Stokes existence and smoothness problem, a complex challenge offering a $1 million bounty. Despite his accusations, Buckmaster finds himself in a paradoxical position: he continues to rely on the very AI models he believes exploited his work.

"Even if you don’t agree with any of this, you’re kind of stuck. With AI being so useful, it’s hard to completely prevent oneself from using it," Buckmaster shared with WIRED, highlighting the pervasive influence and near-monopolistic control of these AI companies. "These companies have a monopoly, and there is not much choice," he added, underscoring the limited alternatives available to researchers.

In the immediate aftermath of his public claims, which thrust the New York University professor into the spotlight, Buckmaster has paradoxically been using OpenAI’s coding agent, Codex, to refine and organize his research papers. When he manages to carve out time for actual mathematical work—a rare occurrence given the media attention—the AI tool has been instrumental in helping him dissect the intricate logical pathways OpenAI’s agents might have traversed to evolve from his preliminary work to the final, complete proof.

Buckmaster’s engagement with AI tools predates the controversy. He had been actively collaborating with Anthropic researcher Levent Alpöge, utilizing both OpenAI’s Codex and Anthropic’s competing Claude model, in his efforts to tackle the Navier-Stokes problem. His contention is that OpenAI, having become aware of the imminent solution, then deployed tens of thousands of its own agents to swiftly arrive at the answer, effectively outpacing him.

Buckmaster’s decision to go public with his allegations ignited a significant debate within the scientific community. It sparked widespread discussions about the future of mathematics and whether AI advancements could render human mathematicians obsolete. The accusations also prompted OpenAI to launch an internal investigation. Subsequently, the company amended its public announcement regarding the Navier-Stokes solution, stating that it "confirmed that Buckmaster’s Codex prompts over the two months preceding this announcement and paper on September 8, 2026, could not have influenced the system in any way, including through training." OpenAI directed WIRED to this revised statement in an email, standing by its assertion.

For Buckmaster, the core issue extends beyond the mere solution of a math problem. He views the demonstration of AI pushing mathematical boundaries as "more important than the result." However, he sharply criticizes the practice of AI companies churning out solutions to long-standing mathematical problems without adequately crediting the foundational human research, especially when such announcements coincide with significant financial events like major IPOs. He labels such actions as "irresponsible" and "childish."

Similar concerns have resonated among other prominent mathematicians. German mathematician Andreas Thom, whose highly specialized techniques in geometric group theory are understood by only a handful globally, found himself in a comparable situation. In August, when OpenAI announced that its Astra model had utilized these very techniques to solve a long-standing problem he had been pursuing for two decades, Thom was "amazed." His immediate question was, "how did they learn about it?"

Thom reached out directly to OpenAI researchers Mark Sellke and Sébastien Bubeck. In an August email, he pointed out that OpenAI’s initial assertion of "no progress" on the problem in the last decade was inaccurate, overlooking his own 2019 paper and the contributions of other mathematicians. Following this, OpenAI amended its press release. Thom and a colleague had been using ChatGPT to aid their work on the problem in the months leading up to OpenAI’s result. When he inquired if their interactions had been incorporated into the AI’s training data, Sellke reportedly replied, "That did not happen."

Despite his skepticism, Thom chose not to pursue the matter further. "I set it aside," he recalled. "I’m not so much interested in these political things; I want to work on mathematics." While aware of OpenAI’s statement regarding Buckmaster’s prompts, Thom admits he doesn’t fully trust it and believes he may never truly ascertain whether his own work inadvertently contributed to OpenAI’s breakthroughs. "AI really kills this entire idea that you could trace back who contributed what," he lamented. "That is probably over."

This inability to trace intellectual contributions marks a radical departure from the established norms of scientific progress. Traditionally, academic research thrives on peer review, transparent attribution, and the collaborative building upon each other’s work with explicit credit. Beyond this erosion of attribution, the inherent opacity of AI’s problem-solving processes—where humans often cannot fully comprehend the precise logical steps taken by the agents—poses profound existential questions for mathematicians. Their field, once meticulously constructed through human insight, appears to be slipping beyond their complete understanding.

"If I want to make a contribution to mathematics, how do I do that as just a human nowadays when these trillion-dollar companies are in on the game?" questioned Cornell mathematician Alex Townsend, co-author of an upcoming book on the evolution of the field. Since August, Townsend has observed a noticeable shift among his colleagues, many of whom are now actively exploring the technology, seeking to understand its intricacies and secure subscriptions to powerful AI models.

Townsend articulates a common sentiment of ambivalence: "I feel both excited and nervous simultaneously. Excited because I can achieve things that I couldn’t achieve without it, and nervous because I’m questioning: ‘OK, what’s my purpose here?’"

Thom, while accepting the evolving landscape of his study, continues to utilize ChatGPT, albeit with updated privacy settings to prevent his work from being used for further model training. He finds it "extremely efficient" for accelerating paper writing. (It’s worth noting that OpenAI’s enterprise and university offerings typically default to not training models on user data). He draws a crucial distinction: "If a human had actively done that, then I would be very, very angry," referring to the unauthorized use of his work. However, if information was integrated into a model via an opaque algorithm that nobody fully understands, "I could probably live with that," he concedes.

Not all mathematicians share Thom’s pragmatism. A powerful statement of concern came from twenty-five Fields medalists, who co-authored an open letter asserting that the interests of AI companies and mathematicians are "severely misaligned." This sentiment is echoed in the Leiden Declaration, which has garnered over 4,000 signatures and proposes a series of recommendations for mathematicians, funders, and policymakers to safeguard the field from being subsumed by AI. Buckmaster articulates a shared fear: the possibility of using AI for minor tasks, like "clean[ing] up some of your grammar, and suddenly your years of work [could be] gobbled up in user data and sold to another mathematician or grad student. That’s what I think most of the mathematicians tend to be worrying about, and I think it’s a real issue."

With no immediate regulatory oversight in sight and AI becoming increasingly entrenched in academic research, some mathematicians are advocating for a temporary slowdown. Over 2,000 individuals affiliated with Caltech called for the suspension of an AI math hackathon at the institution, an event initially sponsored by both Anthropic and OpenAI (though OpenAI has since withdrawn its sponsorship). However, such attempts to decelerate progress may prove futile in the long run.

"I don’t think this is really sustainable because of the efficiency gain" that AI offers, Thom remarked. He believes that mathematicians, particularly early-career researchers, risk becoming "isolated" if they choose not to embrace AI to accelerate their work. Both Thom and Buckmaster agree that the mathematical community must proactively address the implications of this technology for younger generations and facilitate a smoother transition into this new era.

The next generation of mathematicians is already seeking answers. Townsend reports that students are increasingly inquiring about their future career prospects now that AI demonstrates such formidable capabilities in solving complex mathematical problems.

Buckmaster is advocating for a "detente" in AI-driven mathematics, urging mathematicians and AI laboratories to collaboratively establish clear ground rules for the release of research results, including accurate and comprehensive referencing. On a personal level, he is committed to refining the papers he hastily published last week to pre-empt OpenAI’s announcement, one of which he candidly described as "AI slop."

"I have a responsibility to clean up the papers that I did post that weren’t completed, and I think I have a responsibility to explain to mathematicians what we did," he stated. He also expresses openness to engaging in discussions with OpenAI. "I don’t want to just engage in fights," he affirmed. However, when considering the prospect of directly collaborating with the company on a problem, Buckmaster offered a cautious response, delivered with a wry smile: "we have to be careful with that."