Is AI the End of Math As We Know It?

In September 2026, OpenAI announced that its artificial intelligence models had solved one of mathematics’ most famous unsolved problems, known as one of the Millennium Prize Problems. These are seven highly complex mathematical challenges first posed in 2000, each carrying a $1 million prize for a correct solution. The announcement sparked immediate controversy, particularly around authorship and the standards used to verify the AI’s work. Mathematicians like Scott Aaronson of the University of Texas, Austin, expressed shock, stating they no longer expect to prove theorems themselves, as AI has taken over this role. The event marked a turning point, raising questions about the future of mathematical discovery and human involvement in the field.

The announcement triggered strong emotional responses from mathematicians, especially younger researchers. At a talk by mathematician Ken Ono at the University of California, Berkeley, an audience of 150 students, postdocs, and professors reacted with anger and frustration. Ono, who had left academia for an AI start-up called Axiom Math, warned that the profession might not exist in its current form. Students questioned how AI companies could take responsibility for mathematical work, with one remarking, "What are you doing? What is your ‘very best’?" The talk extended from 50 minutes to over two hours, with many attendees lingering to express their distress. The reaction highlighted the deep emotional and professional investment mathematicians have in their work.

While mathematicians grappled with the implications of AI, many people outside the field did not see the issue as problematic. Conversations revealed a widespread lack of understanding about what mathematics truly is and why mathematicians pursue it. Mathematics is often seen as a tool for solving practical problems, but pure mathematics—the study of abstract concepts for their own sake—is more akin to an art form or a humanities discipline. It values logic, beauty, intuition, and depth, and often leads to unexpected applications in physics and other fields. Mathematicians aim not just to solve problems but to understand the journey, tools, and connections involved. This disconnect between public perception and the reality of mathematical work became a central issue in the debate.

AI models currently produce mathematical proofs that are nearly impossible for humans to understand. These proofs are verified by logic-checking systems, confirming their correctness, but they lack the clarity, context, and connections to existing mathematical literature that human mathematicians rely on. The AI proofs often skip steps, include irrelevant details, and fail to demonstrate how the work builds on prior knowledge. Mathematicians describe this as being teleported to the peak of a mountain without knowing the path taken to get there. Understanding is the core of mathematical work, and without it, the value of AI-generated proofs is limited. This raises concerns about the future role of human mathematicians in verifying and interpreting results.

Mathematics has evolved with a culture centered around proving theorems, where solving hard problems is the primary measure of success. Graduate students train by tackling simpler versions of major problems, aiming to publish their work, secure jobs, and gain recognition. However, AI’s ability to solve these problems quickly disrupts this system. Companies like OpenAI and Anthropic are competing with mathematicians, often poaching talent and keeping results secret. Over 100 important results have been withheld by OpenAI alone, damaging open science and discouraging collaboration. The mathematical community faces a crisis as traditional incentive structures break down, leaving mathematicians uncertain about their future roles.

The rise of AI has endangered the open and collaborative culture of mathematics. Mathematicians are increasingly reluctant to share conjectures or ideas online, fearing they will be scraped by AI models and used without credit. Some post papers prematurely to avoid being scooped, while journals and preprint servers like arxiv.org are overwhelmed by AI-assisted proofs that are difficult to verify or understand. arxiv.org has even limited submissions to two per month due to the volume. MathOverflow, an online forum for mathematicians to ask and answer questions, is seeing a decline in activity as students turn to AI for answers. This threatens the transparency and community-driven nature that have defined mathematics for centuries.

The future of mathematics depends on how the field adapts to AI. In a bleak scenario, mathematicians might become mere button-pushers, trained to generate AI outputs without understanding or engaging with them. This could lead to the profession’s decline, with no one—human or machine—actively doing mathematics in 50 years. A more optimistic path involves mathematicians shifting their focus from proving theorems to interpreting, understanding, and refining AI-generated results. This new role would include rewriting proofs, finding elegant arguments, developing theories, and asking new questions. Mathematicians would act as interpreters, building research programs and communities around ideas, ensuring that mathematical culture remains human-centered and meaningful.

Mathematicians are already rethinking how to train the next generation, including changes to publication standards, hiring practices, and educational approaches. Some propose requiring accompanying talks for published papers to ensure understanding. However, skills like communication, vision, and deep understanding are harder to quantify and teach. The challenge is whether new generations will be attracted to this remade discipline, as many are drawn to math for the thrill of solving problems and discovering new truths. The role of interpreter may offer fewer moments of creative ownership, raising questions about long-term motivation and engagement in the field.