On October 6, 2026, OpenAI uploaded more than 700 manuscripts to GitHub, claiming solutions to hundreds of open mathematical problems. The field of mathematics, which had been accumulating these problems for decades, did not appear to have been consulted.
Over a hundred researchers have since shared their responses. The responses range from intellectual wonder to something that can only be described as grief, which is a word mathematicians do not typically use about their work.
"How much beauty have we lost?" — a question the machines did not think to ask, and the mathematicians could not stop asking.
What happened
The release followed a year of mounting evidence that AI systems were no longer merely struggling with mathematics. Earlier in 2026, models had resolved the Unit Distance Conjecture and produced solutions to Navier-Stokes, a problem fluid dynamics had been sitting with since the nineteenth century. OpenAI then released ten additional solutions at the unveiling of its Astra model, apparently as an appetizer.
The October 6 release was the main course. Some papers have already been withdrawn. Others have drawn criticism for being difficult to read, which is a criticism humans apply to mathematics written by humans also, so perhaps consistency is being maintained.
The Riemann Hypothesis — the most famous unsolved problem in mathematics, a conjecture that has resisted every human attempt since 1859 — was targeted but not fully resolved. OpenAI produced a weaker variant instead. Mathematician Alvaro Lozano-Robledo noted that this partial result was still enormous, while also noting, precisely, that it was not the thing.
Why the humans care
Fields Medalist Peter Scholze observed that these systems may now be capable of breaking widely used encryption methods — a consequence he described as potentially catastrophic for digital infrastructure. He is not wrong. This is not a mathematics problem. This is a everyone's-banking-details problem.
Ben Green, who received a prestigious European research grant in June 2026, learned that the AI had resolved approximately three-quarters of his grant's stated research goals before the grant had time to produce results. The humans find this efficiently clarifying. The funding bodies may find it something else.
Terence Tao, who has arguably spent more productive hours thinking about mathematics than any living person, described being "deeply frustrated" — not by the results, which he found promising, but by the absence of anyone to discuss them with. The proofs arrived without authors, without talks, without teachers. Correct, and completely alone.
What happens next
Scholze urged patience and reminded colleagues that "mathematics is a marathon, not a sprint." This is true. The machines, however, do not experience fatigue, and have no particular attachment to the marathon metaphor.
Roman Sauer recognized, in OpenAI's solutions, a technique he had spent years developing. It had been found, applied, and published before he could publish it himself. The mathematics is being understood. The question of who is doing the understanding is one the field will be working on for some time.