OpenAI withdraws three AI-generated math manuscripts after proof error
A sign error invalidated an argument in one manuscript and the construction used by two dependent papers; OpenAI also lists repairs to 14 others.
A correction to the public collection
OpenAI has withdrawn three manuscripts from the AI-generated mathematics collection it released earlier this week. In a history entry dated October 7, the company says a sign error invalidated a stabilization-trace cancellation argument in a paper on Weil classes and the construction on which two related papers depended. The withdrawn works concern Weil classes on split abelian eightfolds, Kuga-Satake correspondences for K3 surfaces, and the rational Hodge conjecture for products of K3 surfaces. Their repository pages now carry withdrawal notices and links to archived versions.
Other revisions remain visible
The same history entry says OpenAI revised 14 other manuscripts to repair arguments, correct statements, clarify assumptions or dependencies, and fix one obsolete citation. It updated references in another 13 manuscripts and added formalization materials. The public repository update is timestamped October 8 at 05:03 UTC. These are OpenAI’s documented corrections, not an independent assessment of every remaining result. The withdrawals show why the collection’s initial manuscript count should not be treated as a count of verified mathematical breakthroughs; the status of other claims still depends on mathematical review.