Only 300 of the 722 results have been formalised, meaning converted into computer-verifiable form, and the repository already carries several retractions and corrections. Researchers who examined the reports found nonsense explanations, formulas that do not make sense, and work already completed by others. OpenAI did not name the model or release the prompts behind the results, despite recommendations from the Advisory Group on Mathematicians and Artificial Intelligence, though it has published changelogs for the GitHub repository.

The deeper worry is cultural. NYU professor Tristan Buckmaster told Business Insider that the release had upended early-career mathematicians, wiping out whole research programmes and leaving some fearful that their own use of AI tools had inadvertently contributed to the results. Northwestern's Bryna Kra argued that this could make mathematicians more secretive, undermining the collaboration the field depends on. Terence Tao of the University of California wrote on Mastodon that "many fewer seminars, workshops, collaborations, or other activities are being generated from these results compared to traditional breakthroughs," and suggested mathematics should value community building over raw solutions.

There are also security concerns. Johns Hopkins associate professor Matthew Green said on social media that he believes the convergence of mathematical and cybersecurity capabilities in frontier AI could result in systems that defeat encryption as we know it, potentially putting cryptocurrency worth trillions of dollars at risk. Not everyone agrees that outcome is inevitable, and the claim remains a prediction rather than a demonstrated capability.