Math group warns AI could weaken how math research is checked

In short: A group of mathematicians published the Leiden Declaration, warning that AI and growing tech industry influence could undermine how math research is reviewed, credited, and shared.

What happened

A working group of 16 researchers released the Leiden Declaration on Artificial Intelligence and Mathematics on June 2, 2026. The group developed it over eight months after a 2025 conference at Leiden University in the Netherlands. The declaration has gathered hundreds of signatories and has been endorsed by the International Mathematical Union, a major global organization in mathematics.

The declaration lists several risks. It says AI systems can produce arguments that look like real proofs but may be wrong, and that can be hard to spot. A proof is the step by step reasoning that shows a math claim is true, like showing your work in school but at a much higher level. The authors warn this could overload reviewers and lead to published results built on mistakes.

It also raises concerns about credit and ownership. The declaration says AI tools often fail to properly cite the human work they draw from, and that some AI training has used material in ways that may break licenses or copyright.

The timing comes shortly after OpenAI promoted an AI result that it said disproved an 80 year old geometry conjecture. The declaration warns that companies may publicize results through press releases and videos before the usual community checks, and without sharing key details needed for outside evaluation.

Why it matters

Math research underpins things people rely on, from encryption that protects online banking to systems used in transportation and medicine. If incorrect or poorly credited work spreads, it is like building a house on a shaky foundation, problems can show up much later and be hard to fix.

Source: Arstechnica