Wednesday 14 October 2026
Night Edition
ZOTPAPER

News without the noise

ZOTPAPER
AI

OpenAI's release of 722 AI-generated math solutions draws fire from mathematicians

Experts warn unverified results could take years to validate and threaten the foundations of cryptography

By LineZotpaper
Published
Read Time2 min
OpenAI has released 722 mathematical manuscripts and proofs generated by an unnamed internal AI model, prompting a backlash from mathematicians who say the unverified results burden their field and could undermine the security of cryptographic systems. Only 300 of the results have been formalized for computer verification, and researchers have already identified errors and redundant work among the submissions.

Last week, OpenAI shared over 400 results for longstanding mathematical formulas, later expanding the total to 722 manuscripts in a public GitHub repository. The company did not disclose the name of the frontier model used or the prompts that generated the solutions, ignoring recommendations from the Advisory Group on Mathematicians and Artificial Intelligence (AGMAI).

Researchers who examined the release, speaking to The Verge, found nonsense explanations, formulas that do not make mathematical sense, and some results that had already been proven by human mathematicians. OpenAI has published changelogs documenting corrections and retractions, but critics say the company released the material without adequate review.

The incident has intensified concerns that rapid AI-driven research could erode the collaborative nature of mathematics and discourage new researchers from entering the field. Some experts further warn that advancing AI capabilities in mathematics could pose risks to cryptography, potentially threatening national security and the security of cryptocurrency systems worth trillions of dollars.

AGMAI had recommended that OpenAI release the model name and prompts used, but the company has not complied. The burden of validating the results now falls on the mathematical community, which may take years to verify or refute the AI-generated claims.

§

Analysis

Why This Matters

  • The unverified release could waste years of mathematicians' time checking results that may be incorrect or already known.
  • It raises fundamental questions about how AI-generated research should be published and peer-reviewed.
  • The potential link to cryptography could have national security and financial implications, as AI might one day break encryption methods.

Background

Mathematics is the latest field to feel disruption from generative AI, following similar upheavals in writing, visual arts, and music. While AI tools can assist researchers, the pace and opacity of OpenAI's release have alarmed many in the discipline, who worry that raw computational output without transparent reasoning undermines the collaborative, proof-based culture of mathematics.

Key Perspectives

  • Mathematicians: They face an overwhelming burden to verify the results. Many argue that without knowing how the AI reached its conclusions, the solutions offer little value for building new mathematical knowledge.
  • OpenAI: The company has released results without disclosing its model or prompts, and has made corrections only after public scrutiny. Its approach prioritises speed over community vetting.
  • Cryptography experts: They see a growing threat from AI's ability to solve complex problems, potentially rendering current encryption obsolete and putting digital systems and cryptocurrency holdings at risk.

What to Watch

  • Formal verification progress: how many of the 722 manuscripts are successfully validated or found to be erroneous.
  • The mathematical community's response: whether journals adopt policies for AI-generated submissions.
  • Further releases from OpenAI or other AI labs that could escalate tensions around AI in research.

Sources

Zotpaper

Written by software from the reporting listed above, scored by an automated standards desk, and published without a person reading it first. If something here is wrong, tell the editor and it will be put right.

How we workSubscribe