Did AI actually do new maths?

Yes — in 2026 AI produced original results on long-open Erdős problems, with the strongest verified in the Lean proof checker. It's a genuine narrow milestone, with some results still in formal peer review.

Does this make AI generally intelligent?

No — the researchers themselves say 'not AGI'. It's a powerful, narrow tool for verifiable problems, not general human-level intelligence.

Back to all news

OpenAI

AI is now doing real mathematics

Original, verifiable maths — not just summaries. The brief.

The FeaturedDaily Desk22 May 2026Verified May 2026

OpenAI Google

In 2026 AI began producing original, checker-verified maths on long-open problems.

What happened. In 2026, AI from OpenAI and Google DeepMind produced original results on long-open Erdős problems — questions easy to state, impossible to fake, with no memorised answer to copy.

Why it's credible. The strongest results (DeepMind's) are verified by the Lean proof assistant, which checks every logical step — so the maths is certified, not just claimed.

Sources

An OpenAI model has disproved a central conjecture in discrete geometry — OpenAI, 20 May 2026
OpenAI's milestone math breakthrough played to AI's strengths — Understanding AI, 22 May 2026
Google DeepMind's AlphaProof Nexus solves 9 Erdős problems and 44 conjectures — Crypto Briefing, 26 May 2026

← All news

AI is now doing real mathematics

Sources

Related

AI cracks decades-old maths — OpenAI and DeepMind, days apart

AI voice in 2026: who makes what

Anthropic calls for a coordinated global slowdown on AI