Mathematicians Depend on AI Tools Despite Attribution Concerns

The mathematics community faces an uncomfortable paradox: researchers believe AI companies may be exploiting their work without proper credit, yet they cannot stop using the very tools that threaten their field's traditional norms.

New York University mathematician Tristan Buckmaster suspects OpenAI used his approach to solve the Navier-Stokes existence and smoothness problem—a legendary challenge with a $1 million prize—before he could complete his own proof. Despite this, Buckmaster continues using OpenAI's Codex to refine his research papers and understand the logical steps the company's agents took to reach their solution.

"Even if you don't agree with any of this, you're kind of stuck," Buckmaster told WIRED. "With AI being so useful, it's hard to completely prevent oneself from using it."

A pattern of disputed attribution

Buckmaster is not alone in his concerns. German mathematician Andreas Thom spent two decades developing techniques in geometric group theory that only a handful of people worldwide understand. When OpenAI announced in August that its Astra model had used those methods to prove a long-standing problem he was working on, Thom questioned how the system learned about his specialized approach.

Thom had been using ChatGPT to assist his work in the months before OpenAI's announcement. When he asked company researchers whether his interactions had been incorporated into training data, he was told they had not. OpenAI later amended its press release to acknowledge Thom's 2019 paper and other mathematicians' contributions, which the original announcement had overlooked.

Following Buckmaster's public accusations, OpenAI conducted an investigation and stated that Buckmaster's Codex prompts "could not have influenced the system in any way, including through training." However, Thom remains skeptical, saying he will probably never know whether his work actually fed the result.

Why it matters

The tension between AI's productivity benefits and concerns about intellectual property represents a fundamental shift in how mathematical research operates. Traditional academic practices of peer review, attribution, and building transparently on others' work are colliding with opaque AI systems that can synthesize vast amounts of information without clear provenance. For an industry built on rigorous proof and credit, the inability to trace contributions threatens core professional values—yet the competitive pressure to use AI tools may be too strong to resist.

The efficiency trap

Cornell mathematician Alex Townsend describes the community's dilemma: "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?"

Since August, Townsend has observed many colleagues asking how to access higher-powered models. The technology offers clear advantages—Thom calls it "extremely efficient" for speeding up paper writing, even as he worries about attribution.

The competitive dynamics may leave mathematicians little choice. Thom warns that researchers who avoid AI risk becoming "isolated," particularly early-career mathematicians who need to publish to advance. "I don't think this is really sustainable because of the efficiency gain," he said.

Calls for ground rules

More than 4,000 people have signed the Leiden Declaration, which offers recommendations for how mathematicians, funders, and policymakers can prevent AI from overwhelming the field. Twenty-five Fields medalists wrote in an open letter that AI companies and mathematicians are "severely misaligned."

Buckmaster is calling for a pause in AI-driven mathematics while researchers and laboratories establish ground rules for releasing results, including proper attribution. He remains open to discussing these issues with OpenAI, though he acknowledges the need for caution.

The details were first reported by WIRED.