What happened?
OpenAI published 722 manuscripts and proofs aimed at mathematical problems that have resisted solution for decades. Only 300 of the results were "formalized" — that is, converted into a computer-verifiable form. That makes it hard to tell which results constitute genuine research.
Why mathematicians are upset
Researchers found nonsensical explanations, inconsistent formulas, and work already completed by others. OpenAI did not disclose the name of the model that produced the results or the prompts used, despite recommendations from the Advisory Group on Mathematicians and Artificial Intelligence (AGMAI). The company only published changelogs in its GitHub repository; several retractions and corrections have already followed.
Impact on the field
NYU professor Tristan Buckmaster said the release upended the research programs of many early-career mathematicians. Northwestern University professor Bryna Kra worries about mathematics' collaborative nature: researchers becoming more secretive could reduce the sharing of ideas and data.
A broader risk
The debate is not only academic. Experts warn that the growing overlap between frontier AI's mathematical and cybersecurity capabilities could put the reliability of cryptography — and trillions of dollars in crypto assets — at risk.
