Mathematicians Dislike AI, Yet Find It Unavoidable

Mathematician Tristan Buckmaster claims that OpenAI utilized his research to swiftly advance and potentially outpace him in solving a renowned mathematical problem with a $1 million prize attached.

Yet, this hasn’t deterred him from employing the company’s models—and he isn’t the only mathematician with similar sentiments.

“Even if you disagree with this, you’re in a tough spot. With AI’s utility, it’s challenging to completely avoid using it,” Buckmaster tells WIRED. “These firms hold a monopoly, leaving limited options,” he adds.

Since the New York University professor publicly accused OpenAI of mimicking his methods, Buckmaster has been leveraging the company’s coding agent Codex to refine his research papers. Whenever he finds time to engage in mathematics (which he notes is rare due to the sudden attention), the tool aids him in piecing together the logical paths that OpenAI’s agents may have followed to arrive at the final proof from his earlier work.

Buckmaster has used both Codex and Anthropic’s rival Claude to tackle the Navier-Stokes existence and smoothness problem, collaborating with Anthropic researcher Levent Alpöge. According to Buckmaster, OpenAI deployed tens of thousands of agents to achieve the solution only after discovering that the equation was nearing resolution.

Buckmaster’s public claims sparked intense discussions about artificial intelligence and its potential to render human mathematicians obsolete. This prompted OpenAI to conduct an investigation and revise its announcement regarding the solution to Navier-Stokes, clarifying that it “confirmed that Buckmaster’s Codex prompts over the two months prior to this announcement and paper on September 8, 2026, could not have influenced the system in any way, including through training.” The company referred WIRED to its announcement via email.

Demonstrating that AI is expanding the limits of mathematics was “more significant than the outcome,” Buckmaster remarks. However, producing solutions to long-standing mathematical challenges without giving due credit to the human contributions behind them—especially ahead of significant IPOs—is irresponsible and “childish,” he states.

Other mathematicians share similar apprehensions. Very few individuals worldwide grasp the techniques in geometric group theory that German mathematician Andreas Thom has devoted two decades to developing. Thus, when OpenAI announced in August that its Astra model had utilized these techniques to prove a problem he has been addressing, “I was astonished,” says Thom. “Naturally, I wondered how they came to know about it.”

He subsequently reached out to OpenAI researchers Mark Sellke and Sébastien Bubeck. In an August email, he pointed out that the company’s claim of “no progress” on the problem in the last decade overlooked a 2019 paper of his, among other mathematicians’ contributions. The company updated its press release accordingly. He and a colleague had been utilizing ChatGPT to support their work on the problem in the months leading up to the result, but when he inquired whether their interactions had been incorporated into the training data, Thom reports that Sellke responded: “That did not happen.”

“I let it go,” Thom reflects. “I’m not particularly interested in these political issues; I want to focus on mathematics.”

While Thom has encountered OpenAI’s statement asserting that Buckmaster’s prompts couldn’t have influenced the system, he admits he doesn’t trust this and concedes that he may never find out if his work truly contributed to the result.

“AI fundamentally undermines the notion that you can trace back who contributed what,” he asserts. “That idea is probably gone.”