AI progress raises alarm among top mathematicians

Tao Zhexuan warns: AI is exhausting "good problems" in mathematics

Tao Zhexuan, a professor at the University of California, Los Angeles (UCLA) and 2006 Fields Medal winner, issued a warning on the Mathstodon website on September 8, 2026, which was widely circulated in academia. He pointed out that artificial intelligence is no longer just a tool for researchers, it is threatening human understanding and the nature of basic research. In fact, artificial intelligence is depleting the resource pool of "high-quality open questions." These unresolved issues that truly promote progress in the scientific field are at risk of disappearing due to the intervention of AI.

Summary of Core Views

• AI can solve long-term problems in a matter of days that humans will take years to overcome.
• Tao Zhexuan warned that AI is draining the reserves of "good problems" in the mathematical world.
• Mathematicians worry that the value of their roles and discoveries may be devalued.
• Academic circles are calling for the development of new rules to regulate the use of AI in research.

AI solves millennium problems at record speed

Since May 2026, announcements about artificial intelligence have been continuous. OpenAI first denied the Erdös conjecture about unit distance, an open question that has been unresolved since 1946. A few weeks later, the same AI company claimed to have solved one of the seven "Millennium Award Questions" of the Clay Institute of Mathematics: the existence and regularity of solutions to the Navier-Stokes equation. These equations are used to describe the motion of fluids.

According to Tristan Buckmaster, he is a promising author of research methods, and OpenAI's AI completed in one weekend what he and his colleague Levante Alpager (Anthropic employee) had been trying to perfect for months.

But that's not all! In August 2026, OpenAI also released ten major mathematical and computer science results achieved by its Astra model. These results cover geometry, cryptography and coding theory. At the same time, Anthropic used Claude to formalize the proof of Fermat's Last Theorem in just 11 days. The AI company just announced that it is about to launch an initial public offering (IPO).

Tao Zhexuan: AI is running out of "good problems" in mathematics

Tao Zhexuan, known as the greatest mathematician alive, issued a clear warning in a post posted on Mathstodon. He believes that the real danger is not that AI can solve problems, but that it solves problems so quickly that the community has not had time to learn all the lessons from it.

He wrote: "Indiscriminate use of powerful solution-extraction tools may achieve the immediate goal of solving problems, but at the cost of sustaining the ecosystem for the next wave of progress." (Inselective use of powerful problem-extraction tools may achieve immediate goals of solving current problems, but at the expense of undermining the sustainability of the ecosystem for the next wave of progress.)

For Tao Zhexuan, the value of a mathematical problem lies both in its solution and the path taken to reach the solution. Indeed, this process can:

• discover new methods;
• find new connections between different domains (sometimes); and
• reconstruct the problem itself.

However, with AI, this path is bypassed. In this regard, Tao Zhexuan warned: "We have now seen that even rumors that someone is studying a certain problem can trigger a massive amount of AI-driven efforts to 'flat' the original research project before it reaches its full potential."

Explanation: AI transformed mathematics from a subject of "proof scarcity" to a subject of "proof excess". This may lead to the devaluation of human work.

Faced with AI, mathematicians face identity crisis?

In addition to Tao Zhexuan's remarks, the entire professional field is reflecting on itself. "If the main goal of mathematics is to prove theorems, it's easy to ask: Since machines seem to be able to do this at will, what use are we doing?" asked Henry Yuen, a mathematician at the University of Colombia, in an interview with the Daily Telegraph.

In June 2026, more than 3000 mathematicians signed the Leiden Declaration. The document calls for:

• to use AI responsibly;
• to strictly verify results; and
• to appropriately cite the contributions of humans and artificial intelligence.

Tao Zhexuan himself suggested adopting an approach: classifying certain issues into issues that require in-depth analysis. This means that a raw answer (especially one provided by artificial intelligence) only counts if it is accompanied by an understandable and instructive reasoning process.

Where is the future of mathematics in the AI era?

Some people view AI as a productivity tool. The truth is, it frees humans from cumbersome tasks and allows them to focus on asking new questions. On the other hand, there are also concerns about the industrialization of the certification process. They believe that intellectual value may be faded by the pursuit of speed.

Both sides agreed on one thing: the rules must change. The Leiden Declaration therefore emphasized transparency and recognition of contributions. But how to apply these principles in a competition where AI models are not public and certificates are generated in a matter of hours?

Regardless, AI has crossed a threshold in mathematics. Now the ball is on the community's side: Defining new rules to ensure that artificial intelligence remains a tool rather than a substitute.