A growing dispute between OpenAI and the mathematics community is highlighting a challenge that extends far beyond academic research: when artificial intelligence contributes to a discovery, who deserves credit, how should the result be verified and what happens to the culture of open collaboration?

Twenty-five Fields Medal recipients have signed an open letter warning that competition between leading AI companies could threaten the way mathematical knowledge is created, shared and validated. The Fields Medal is widely regarded as one of the most prestigious distinctions in mathematics.

The signatories acknowledge that AI systems could help solve some of the world’s most difficult mathematical problems. However, they argue that a solution has limited value if researchers cannot understand it, verify it and integrate its methods into the wider body of mathematical knowledge.

A Controversy Over Attribution

The debate intensified after Tristan Buckmaster, a professor at New York University, accused OpenAI of pressuring him not to credit a collaborator employed by Anthropic for contributing to the solution of an important mathematical problem.

Buckmaster also questioned whether OpenAI might have used insights from the researchers’ work with Codex to generate its own proof during an intensive inference process. OpenAI’s proof has yet to be independently verified.

The controversy has generated broader concerns among researchers about how interactions with AI tools may be stored, reused or incorporated into future model development. For mathematicians, the uncertainty creates a difficult dilemma: using advanced AI systems may accelerate their research, but it could also expose promising ideas before they are ready to be published.

OpenAI later withdrew its sponsorship of a mathematics event at Caltech following criticism from researchers at the institution, further escalating tensions between the company and parts of the academic community.

Why Verification Still Matters

Mathematical discovery involves more than producing a correct final answer. Researchers must document the reasoning behind a proof, identify the techniques it introduces, connect it to previous work and allow other experts to examine every step.

According to the open letter, AI-generated solutions are sometimes announced before this process can take place. The rush to demonstrate progress may leave insufficient time for rigorous documentation, independent verification and appropriate attribution.

This creates both technical and ethical risks. A proof that cannot be clearly explained or checked remains difficult to trust, regardless of the computing resources used to produce it. At the same time, failing to recognise the human work behind the result raises questions about plagiarism and intellectual ownership.

The mathematicians argue that human expertise remains essential for turning an AI-generated idea into durable knowledge. Researchers must interpret the result, refine its arguments, explain its importance and teach the underlying concepts to future generations.

Could AI Competition Encourage Secrecy?

Mathematics has traditionally benefited from a relatively open research culture. Scholars share partial results, discuss possible approaches and build on one another’s ideas.

Frontier AI laboratories, however, have access to levels of computing power that individual researchers and universities cannot match. If a promising research direction becomes visible, an AI company could invest significant resources in attempting to reach the final proof first.

Even the perception that this might happen could change researchers’ behaviour. Mathematicians may become more reluctant to discuss unfinished work, collaborate openly or test sensitive ideas through commercial AI platforms.

The result could be a less transparent scientific environment, even if the technology itself makes some forms of discovery faster.

A Broader Framework for AI-Assisted Mathematics

The open letter follows the Leiden Declaration, published by a working group of mathematicians in June. That initiative examines how large language models could reshape mathematical research and proposes recommendations for researchers, academic institutions and policymakers.

Both documents emphasise that the objective should not be to prevent the use of AI in mathematics. Instead, the priority is to establish standards that protect attribution, support independent verification and preserve the human systems through which knowledge is developed and transmitted.

Possible principles include transparent disclosure of AI involvement, clear rules governing research data, rigorous review of machine-generated proofs and appropriate recognition of everyone whose work contributed to a result.

The Lesson for Other Industries

The debate reflects challenges already emerging in software engineering, scientific research and creative industries.

AI can accelerate the production of code, designs, analyses and new ideas. Yet speed alone does not guarantee quality or progress. Organisations still need people who can validate outputs, understand their origins, document the reasoning and connect individual results to wider objectives.

For technology leaders, the lesson is clear: governance must evolve alongside capability. Companies adopting AI should establish policies covering data use, intellectual property, attribution, human review and transparency before disputes arise.

The conflict in mathematics may appear highly specialised, but the underlying questions are universal. As AI changes how knowledge and intellectual work are produced, every organisation will need to decide what responsible collaboration between humans and machines should look like.

Source