AI-driven breakthroughs in mathematics

In May 2026 an OpenAI model produced a proof that resolved the decades‑old unit distance conjecture. Human mathematicians quickly built on the AI‑generated technique to settle the related sum‑product conjecture, prompting a wave of research that incorporates AI tools alongside traditional methods.

Within weeks, other AI systems, notably from DeepMind, delivered proofs for nine additional Erdős problems, and by July a separate model had established the long‑standing cycle double cover conjecture. Shortly after, mathematician Levent Alpöge used an AI‑derived polynomial map to refute the Jacobian conjecture in dimensions above two.

These rapid proof‑generation advances highlight AI’s capacity to produce formal results far faster than human effort alone. Nevertheless, experts such as Terence Tao note that the process of understanding, contextualising, and integrating these proofs—what they call “digestion”—has not kept pace, creating a mismatch between the volume of AI‑generated proofs and overall mathematical progress. Most new mathematical papers continue to be authored by humans, indicating an ongoing collaborative dynamic rather than a replacement of traditional research.

Overall, the period from May to July 2026 shows AI moving from a single landmark breakthrough to a series of high‑profile conjecture resolutions, reshaping collaboration patterns while leaving the deeper interpretive work firmly in human hands.