OpenAI withdraws three math manuscripts after a shared proof error

OpenAI’s mathematics repository records the withdrawal of three manuscripts after a sign error invalidated an argument and the construction used by two dependent papers. It also lists revisions to fourteen other manuscripts, including proof repairs and clarified assumptions. The update shows why checking a result’s dependencies matters alongside checking its individual steps.

OpenAI · Oct 7, 2026

Why it matters

Engineers evaluating AI research claims should follow revision histories instead of treating an initial release as settled evidence.

More context

Read the affected papers’ notices and archived versions before relying on their claims. The original release includes some Lean formalizations, but the presence of formalized proofs does not establish that every manuscript or supporting assumption has been verified. This is a research correction, not a model release.

Keep in mind

The repository documents these specific withdrawals and revisions. It does not establish that all remaining results are wrong or independently confirmed. We have not checked the proofs and make no claim to resolve their mathematical validity.

Original read