OpenAI Retracts 3 AI Math Manuscripts After Sign Error Triggers Chain Reaction

On the afternoon of October 9, 2026, OpenAI retracted three previously released mathematics manuscripts after a sign error in one paper led to an incorrect key conclusion, which in turn affected two other manuscripts dependent on it.

Just three days earlier, OpenAI had posted 722 manuscripts on GitHub, covering proofs and research related to 372 mathematical problems across 17 fields including number theory, algebra, geometry, and topology. The release drew strong reactions from the mathematics community, with some researchers impressed by the AI's problem-solving ability and others concerned that many proofs were difficult for humans to understand or verify.

The three retracted manuscripts were part of that release. The retraction was triggered by a sign error in one paper that invalidated its central result; the other two papers relied on that result and were therefore also withdrawn. Mistakes in mathematical proofs are not unusual, but the scale and speed of AI-generated manuscripts may amplify the impact of such errors.

Earlier, Terence Tao and 25 other Fields Medalists had issued a joint statement urging AI companies not to treat mathematical problems merely as benchmarks. After this incident, the Human Mathematics Association (AHM) called on mathematicians to stop cooperating with OpenAI, saying its approach disregards research norms. Debate continues over the reliability and academic standards of AI-generated mathematics.