A sudden deluge of artificial intelligence research has upended the academic mathematics community after OpenAI published more than 700 manuscripts outlining proofs, solutions, and progress across theoretical disciplines, according to reports from The Verge and Scientific American.
The release, published Tuesday on GitHub, addresses nearly 400 open problems spanning combinatorics, geometry, number theory, theoretical computer science, and mathematical physics. The sheer volume of material proved so vast that OpenAI included specialized navigation guidance for its repository, leaving researchers sifting through dozens of pages of abstracts just to understand what had been published.
Breakthroughs on classic problems
Despite deep skepticism about presentation and formatting, several specialists told Scientific American and The Verge that parts of the release appear to contain groundbreaking work. Stanford University mathematician Jared Duker Lichtman identified several results that could rival Fields Medal-level achievements, including a solution to the four-dimensional Kakeya conjecture and notable steps toward the famous Riemann hypothesis.
According to Scientific American, the papers claim to settle the question of “Siegel zeros” and prove the “quasi-Riemann hypothesis,” which mathematician Hector Pasten of the Pontifical Catholic University of Chile called “way more than enough for many applications,” including estimating prime number distribution. The release also claims an extension of Hilbert’s 10th problem to rational numbers, proofs related to interacting Bose-Einstein condensates, and an improved theoretical speed limit for matrix multiplication, shrinking the operations exponent to 2.25.
Verification concerns and missing rigor
While theoretical accomplishments drew excitement, researchers emphasized that the vast majority of findings remain unverified. Only about 300 top-line results across 719 manuscripts—roughly 42 percent—were encoded in Lean, an interactive proof assistant software that allows computer verification of mathematical claims, The Verge reported. OpenAI confirmed in repository notes that its manuscripts stand at differing stages of verification.
Outside scientists have already pointed out errors, leading OpenAI to retract three papers. Kevin Buzzard, a professor of mathematics at Imperial College London, told The Verge that very few highlighted results in his field of algebraic number theory were formally verified, warning that scholars must either read potentially incorrect material or wait for third-party formalization. Other mathematicians noted unusually brief arguments, sparse bibliographies, and inconsistent attribution throughout the papers.
Disruption across academic departments
The abrupt publication has caused widespread anxiety across university campuses. Several researchers reported that long-planned academic programs and existing research agendas have been derailed overnight.
“I have a bunch of friends and colleagues whose grant proposals have just been wiped out,” Colva Roney-Dougal, a mathematics professor at St Andrews University, told The Verge.
Disruption appeared concentrated in probability, combinatorics, and theoretical computer science. Tristan Buckmaster, a mathematician at New York University, told The Verge he knew of three researchers whose entire research tracks had been obliterated by the publications. In some subfields, however, specialists reported their specific areas remained largely untouched.
Mathematicians say it will take years of painstaking peer review to confirm how many of OpenAI’s proofs are valid and whether the models have genuinely solved problems that have resisted human intellect for generations.
