An Anthropic mathematician using Claude claims to have proven the Riemann hypothesis, one of mathematics' most famous unsolved problems. The claim joins recent AI breakthroughs, including OpenAI's work on the planar unit distance problem and an apparent solution to the Navier-Stokes problem, a Clay Mathematics Institute Millennium Prize Problem.

These advances raise a challenge for how mathematics evaluates and rewards expertise. As AI systems generate proofs at unprecedented speed and scale, the bottleneck in mathematics has shifted from production to judgment. Human mathematicians must now distinguish between correct but trivial results and those that genuinely advance understanding.

The article argues that journals, universities, and funders need to adapt. Journals should require authors to explain what is genuinely new and why it matters, with reviewers judging whether results deepen understanding rather than merely verifying correctness. Universities should stop treating publication counts as proxies for quality, since AI has dramatically reduced the time needed to produce sophisticated mathematics.

Proofs generated by AI may be valid without being illuminating or comprehensible. Some proofs may be so long or opaque that checking them requires substantial specialist work. The job of human mathematicians will be to scrutinize correctness, extract underlying ideas, and determine whether results meaningfully advance the field.

Mathematical training must continue to reward understanding through seminars, discussions, and collaborations, where someone can demonstrate they understand an argument and can connect it to other ideas. This becomes more important when papers and applications can be written mostly using AI.