AI has drawn a great deal of attention for reportedly solving difficult mathematical problems.

Some critics argue that the recent AI solution of the Navier-Stokes problem may have drawn heavily on solutions posted by human mathematicians. But there is a larger concern.

In an open letter, in the form of a Declaration dated September 11, 2026, leading mathematicians — including 17 Fields Medalists — warn that a rush to produce answers could come at the expense of the careful explanation and shared understanding that make mathematical discoveries valuable. The Fields Medal is often regarded as the closest equivalent to a Nobel Prize because there is no Nobel for mathematics.

Their letter, “A Severe Misalignment of AI in Mathematics”—which others are invited to sign— concedes that AI has become much better at solving math problems. But, they argue, getting an answer is only one part of doing mathematics. A solution matters most when people can understand how it works, learn from its methods, and use those ideas to make further discoveries.

When AI companies like Anthropic and OpenAI, race to announce solutions as proof of their systems’ abilities, they may leave too little time to explain the work carefully, credit earlier researchers, or help students and mathematicians learn from it. AI could become a valuable tool for mathematical research, but only if the people developing and using it focus on building understanding rather than producing answers as quickly as possible.

The authors see their letter as part of a larger question for science, the arts, and other fields: as AI changes how work gets done, how do we preserve the learning, creativity, and human connections that make the work valuable?

AI might solve a math problem like Fermat’s Last Conjecture but it doesn’t understand the proofs. On the most fundamental level, AI can add the numbers 22 and 13, but doesn’t understand what the number 22 or 13 mean.

Read the entire Declaration here.