OpenAI releases 700+ AI-generated math manuscripts claiming solutions to hundreds of open problems; mathematicians react with shock and concern

On 6 October 2026, OpenAI uploaded more than 700 manuscripts to GitHub. They claim solutions to hundreds of open maths problems, including "quasi-RH", a weaker variant of the Riemann Hypothesis. They do not claim to have proved the Riemann Hypothesis itself. Some papers have already been withdrawn, others are criticised as hard to read, and no authors or failure rates were disclosed. None of the claimed proofs has been independently verified, and the coverage we hold (The Decoder, four days later) is a secondary reaction piece with no primary links extracted.

If even a modest share of the proofs holds up, the way mathematicians work, publish and referee would change quickly. Researchers, journal editors and early-career academics are most exposed, because checking a flood of machine-written papers is a bigger bottleneck than producing them. The release also tests whether AI output can be verified at scale, which matters for any field that relies on expert review. The real impact is still uncertain until the proofs are checked, and the withdrawals show that at least some claims did not survive first contact.

"Possibly the most important day in the history of mathematics" is one researcher's opinion, not a finding, and it comes before any verification. The encryption-risk remark is speculation: nothing reported shows a result that threatens widely used cryptography, and a weaker variant of the Riemann Hypothesis is not the Riemann Hypothesis. Emotional reactions ("shock and disgust") tell us how mathematicians feel, not whether the proofs are correct, and the missing hit rate means nobody can yet say how many of the 700 are right.

1. 01Independent verification: how many of the 700+ manuscripts are confirmed correct, formally checked (for example in Lean) or refuted, and how many more are withdrawn, within the next 4 to 8 weeks.
2. 02Disclosure from OpenAI: authorship, the process used, the failure rate, and which system produced the work, plus any statement on how the papers will go through journals or arXiv review.
3. 03Responses from journals, arXiv and bodies such as the AMS on policy for AI-generated submissions, and verdicts from named experts like Tao, Green and Scholze on specific results, especially quasi-RH and any cryptography-relevant claims.