OpenAI pulled three papers from its public math research repository, and it hasn't said why.
The removals surfaced in the change log of OpenAI's math GitHub repo, the same file where the company tracks updates to its research output. No announcement, blog post, or press release accompanied the edits. The thread documenting the change showed up on Hacker News and picked up 228 points and 177 comments, with commenters speculating about everything from quiet corrections to more serious retractions. OpenAI has not issued a public statement explaining what was wrong with the papers or detailing when the withdrawal happened.
OpenAI's mathematical capability claims, including past assertions about olympiad-level problem solving, have drawn scrutiny before, and researchers want receipts, not silence. Pulling papers through a git diff rather than a formal retraction notice sidesteps the kind of accountability academic publishing normally demands. Without specifics on which results were affected, outside researchers cannot tell whether this is a typo fix or a correction to benchmarks other teams have already cited.
Academic journals make you publish a retraction notice explaining what went wrong; a changelog commit carries no such obligation, and that gap is doing a lot of work here.