AI shocks mathematicians with hundreds of new proofs

Artificial intelligence is changing mathematics at a pace that has left many mathematicians amazed and deeply uneasy.

OpenAI recently released solutions to 372 previously unsolved problems across number theory, algebraic geometry, topology, and other areas of math. Mathematicians are reeling as AI models solve problems that researchers have spent years trying to crack.

The release follows a turbulent month for AI and mathematics. OpenAI previously announced that one of its advanced models had solved a notoriously difficult Millennium Prize Problem that carries a $1 million prize for whoever cracks it.

Last month, the company said that it had cracked the long-standing Navier–Stokes problem. Recently, it announced through a blog that the tech giant’s AI models have solved more problems, and its 722-page release detailed progress on three of the five Millennium Prize Problems.

Questions arise
The announcement initially looked like a breakthrough, but mathematicians quickly questioned whether the result truly met the standards of a human mathematical proof.

The criticism grew as AI companies increasingly used difficult math problems to demonstrate the power of their latest models. More than two dozen Fields Medal winners warned that the industry’s race could weaken the role of mathematicians and change how discoveries are valued.

Academicians also worry these models may have absorbed their research during training, making it hard to tell whether AI is creatively solving complex problems or simply plagiarizing human work faster.

OpenAI has tried to rebuild trust. The company helped create an independent advisory group of mathematicians and used its recommendations when publishing the latest batch of results.

However, the same group has also warned that the race to publish research at breakneck speed could damage the field and has urged the company to stop solving math problems.

For mathematicians, the challenge is no longer simply whether AI can solve hard problems. It is whether humans can understand, verify, and build on the answers AI produces.