Mathematicians Challenge AI to Prove Unsolved Problems

Top academics propose a math exam to test AI's ability to solve research-level challenges.

Published on Feb. 9, 2026

Got story updates? Submit your updates here. ›

A team of leading mathematicians has proposed a new challenge called "First Proof" to test the abilities of artificial intelligence systems in solving unsolved math problems. The exam will feature a set of research-level math problems that cannot be found in the training data of large language models, forcing the AI systems to demonstrate their ability to tackle novel mathematical challenges. This comes as the AI industry has made claims of proving math results without transparency, raising concerns among academics about the legitimacy of these achievements.

Why it matters

The race to develop AI systems capable of doing pure mathematics has intensified, with startups and tech giants making bold claims about their AI's prowess. However, many mathematicians are skeptical of these claims, arguing that the problems solved so far have not been true research-level challenges. The First Proof exam is an attempt to create a controlled experiment to better assess the current capabilities of AI in mathematics and push the technology to tackle more meaningful problems.

The details

The First Proof exam was developed by a team of eleven prominent mathematicians, including a Fields Medal winner. They have contributed math problems that have arisen in their own research and have encrypted the proofs to the solutions. The AI systems will have one week to solve these problems, and the encrypted proofs will be revealed on February 13th. While the problems are not earth-shattering breakthroughs, solving them would demonstrate the potential of AI to assist mathematicians in the more tedious aspects of their work, such as proving lemmas and smaller theorems.

  • The First Proof exam was announced on February 9, 2026.
  • The AI systems will have one week to solve the math problems.
  • The encrypted proofs to the solutions will be revealed on February 13, 2026.

The players

Andrew Sutherland

A mathematician at the Massachusetts Institute of Technology who was not involved in the development of the First Proof exam.

Daniel Spielman

A professor at Yale University and one of the experts behind the new First Proof challenge.

Got photos? Submit your photos here. ›

What they’re saying

“These are brand new problems that cannot be found in any LLM's training data. This seems like a much better experiment than any I have seen to date.”

— Andrew Sutherland, Mathematician (Scientific American)

“Almost all of the papers you see about people using LLMs are written by people at the companies that are producing the LLMs. It comes across as a bit of an advertisement.”

— Daniel Spielman, Professor (Scientific American)

What’s next

The results of the First Proof exam will be closely watched by the mathematics community to see how well AI systems perform on the research-level challenges. The success or failure of the AI systems in solving these problems will likely shape the ongoing debate about the capabilities and transparency of AI in the field of pure mathematics.

The takeaway

The First Proof exam represents a significant effort by leading mathematicians to create a more rigorous and controlled test of AI's abilities in solving unsolved math problems. This challenge aims to move beyond the hype and claims made by the AI industry and provide a clearer assessment of the current state of the technology in tackling meaningful mathematical research.