On October 8 this digest's Trend Watch asked who would check the 722 manuscripts OpenAI published from an unreleased model. Part of the answer came quickly. The repository's history.md, dated October 7, 2026, records the first withdrawals.
What was withdrawn and why
"In 'Algebraicity of Weil classes on split abelian eightfolds' a sign error invalidates a stabilization-trace cancellation argument and the construction used by two dependent papers." Three manuscripts were withdrawn:
- Algebraicity of Weil classes on split abelian eightfolds
- Algebraicity of Kuga–Satake Correspondences for K3 Surfaces
- The rational Hodge conjecture for products of K3 surfaces
The same entry says OpenAI "revised 14 other manuscripts with proof repairs, corrected statements, clearer hypotheses and dependencies," across topics including Lipschitz heights, Kähler minimal model programs and a Birch–Swinnerton-Dyer formula paper, and updated 13 more to cite revised companions. It added six formalizations, bringing "the total percentage of top-line results formalized to 300 / 719 = ~42%." None of the three withdrawn titles appears in the repository's Lean formalization catalogue; on Hacker News, the first question several readers asked was whether the withdrawn papers had been Lean-checked.
Why the dependency matters
One error in one construction took out two other papers that built on it. That is normal in mathematics, but it usually happens over months, between people who understand the work. Here it happened within a day, in a collection that is itself a dependency graph of hundreds of results. The fixes also show what review cost looks like: 17 of the manuscripts touched in the first round, before most specialists have read any of them.
Tao: from 'Math 1.0' to 'Math 2.0'
On October 6, the day of the release, Terence Tao posted a four-part thread. A traditional breakthrough, he wrote, generates talks, workshops and collaborations through which proofs "become digested, streamlined, placed in context." Now, often, "problems are being solved autonomously by AI prompters who have no interest in the broader field itself once their initial target is 'solved'," with fewer seminars and collaborations following. And "a problem that has been 'solved' cannot be somehow reverted to become 'unsolved'," so open problems "are now being harvested at large scale in an unsustainable fashion." His proposal: "Math 2.0" should "value mathematical progress more holistically," including exposition and community building, and the field must "re-evaluate its criteria for education, publication, and career advancement."
Aaronson: adapt, but keep humans in the loop
Scott Aaronson called the release "surely one of the biggest days in mathematical history," noting it includes a claimed proof of the Unique Games Conjecture with a Lean certificate, while "it also appears that no human has understood just about any of these proofs yet." In an October 9 update he rejected a statement from a group calling itself the Association for Human Mathematics ("Mathematicians did not ask for this work to be done"), calling the position that no one should solve open problems with AI "totally untenable," but argued for keeping "human understanding at the center of our enterprise."
What remains uncertain. How many more results will be withdrawn as specialists read them; whether formalization coverage rises above 42%; and whether the withdrawn statements can be recovered with corrected proofs.
Analysis: the useful lesson reaches beyond mathematics. When a model produces many linked results at once, verification has to follow the dependency graph, not single outputs: one bad shared lemma invalidates everything downstream. Machine-checked proofs catch that; human reading at this volume does not.
OpenAI withdrew three linked Hodge-conjecture papers a day after its math release when a sign error broke a shared construction, revised 14 others, and has Lean proofs for only about 42% of top-line results; Tao and Aaronson agree the open question is human understanding and verification, not production.