Artificial intelligence solves historic math conjectures, sparking new discoveries

In May 2026 an OpenAI model disproved the 80‑year‑old unit distance conjecture, a landmark problem in geometric graph theory. Shortly thereafter DeepMind systems resolved nine additional Erdős problems, and in July a model proved the long‑standing cycle double cover conjecture. On 20 July, mathematician Levent Alpöge posted a short polynomial map that refutes the Jacobian conjecture in all dimensions above two.

These breakthroughs demonstrate that AI can generate proofs far faster than before, yet mathematician Terence Tao notes that digestion – understanding and contextualising proofs – has not accelerated, creating an “impedance mismatch” between proof abundance and mathematical progress. While AI‑driven results are increasing, most new mathematical papers still originate from human researchers, indicating a coexistence rather than a takeover.

The rapid proof generation is reshaping how mathematicians collaborate with AI, with human scientists adapting AI‑derived techniques to further advance the field.