OpenAI AI-generated math papers
What happened
Latent Space reported on September 22, 2026, that OpenAI had announced an independent advisory group of mathematicians to guide assessment of AI math advances.1 On October 6, The Verge reported that OpenAI had published solutions to open math problems from 722 manuscripts on GitHub.2 OpenAI's own report that day said the results came with Lean formalizations, reasoning summaries and compute statistics, following advice from the IAS advisory group.3 Latent Space said on October 7 that the papers came from an unreleased internal model and that the claims were unverified.4
Pushback and withdrawn papers
Latent Space reported on October 8 that the release faced credibility pushback, with three papers withdrawn, 14 revised and proofs not fully Lean-checked.5 The same report said the release's README conceded that issues may exist.5 TechCrunch reported that day that OpenAI followed some principles of AGMAI, the Advisory Group on Mathematics and Artificial Intelligence.67 TechCrunch also said only 10 of 719 manuscripts released chain-of-thought and 42% of proofs had not undergone formalization.6
Zvi Mowshowitz reported 719 manuscripts on October 9, claimed to include 90 of the top 500 open problems, at about three compute hours each.8 The Verge reported that day that OpenAI had released nearly 400 results across more than 700 manuscripts.9 The same report said mathematicians say verification and digestion will take years.9
The Verge reported on October 9 that the three papers were removed over a sign error, which the company said invalidates an argument.10 Mowshowitz reported that day that the three papers were in the unformalized set and that Lean-checked proofs were holding up so far.11
A challenge to the Fourier transform claim
An author at argmin.net argued on October 13 that OpenAI's claimed faster Fourier transform improvement was galactic.12 The author said its hidden constants were so large that it beats Cooley-Tukey only at impossibly large n.12
New results in late October
The Verge reported on October 26 that OpenAI published claimed results in math, computer science and physics, with outside experts guiding or verifying them.13 The Verge said on October 27 that math output was published, computer science explained by an external group and physics deferred to outside experts.14 Latent Space reported on October 28 that OpenAI said its AI invented a technique resolving a 35-year-old wave-focusing problem.15 The same report said the work was not yet peer-reviewed.15
Mowshowitz wrote on October 29 that a result on local smoothing and the Kakeya conjecture was plausibly Fields Medal-worthy.16 Mowshowitz also described an existence proof for certain quantum field theories, related to a Millennium problem.17
Latest reported
The Verge reported on October 30 that Altman acknowledged researcher frustration on a podcast and defended releasing discoveries.18 Altman also said OpenAI has more undisclosed proofs and spent heavily on math problems.18 A second Verge report that day said a claimed Yang-Mills breakthrough had not been released publicly.19 That report said mathematicians and scientists criticized the decision as gatekeeping.19
Sources for the overview
- Latent Space: OpenAI mathematics advisory group formed
- The Verge: OpenAI releases batch of solutions to open math problems
- OpenAI: OpenAI releases new AI-produced mathematical results
- Latent Space: OpenAI releases 722 math manuscripts
- Latent Space: OpenAI 722-paper math release
- TechCrunch: OpenAI releases 719 claimed solutions to hard math problems
- OpenAI: Advisory Group on Mathematics and Artificial Intelligence
- Zvi Mowshowitz: OpenAI releases 719 AI-generated math manuscripts
- The Verge: OpenAI releases nearly 400 AI-generated mathematical results
- The Verge: OpenAI retracts three math papers
- Zvi Mowshowitz: OpenAI withdraws three math manuscripts
- argmin.net: OpenAI claims faster Fourier transform algorithm
- The Verge: OpenAI announces three AI-generated science results
- The Verge: OpenAI announces math, computer science and physics breakthroughs
- Latent Space: OpenAI AI wave-focusing math result
- Zvi Mowshowitz: OpenAI local smoothing and Kakeya result
- Zvi Mowshowitz: Existence proof for gauge and fermion quantum field theories
- The Verge: Altman comments on OpenAI's math results on podcast
- The Verge: OpenAI withholds Yang-Mills theory result details
The stories
- · OpenAI mathematics advisory group
- · OpenAI releases batch of AI-generated math manuscripts
- · OpenAI withdraws three math papers
- · OpenAI faster Fourier transform claim
- · OpenAI math proofs
- · OpenAI math and physics results
- · Astra Standard Model lattice proposal