The statement
The Association for Human Mathematics published a statement on 7 October, posted as a guest piece on Terence Tao’s blog, calling on mathematicians to stop collaborating with OpenAI.
Its central line is about volume rather than content. Releasing more than 700 files at once “is not a demonstration of scholarship, but a demonstration of power,” the group writes. It adds that “mathematicians did not ask for this work to be done”, rejects OpenAI’s claim that the release advances the subject, and urges “mathematicians to discontinue their work with OpenAI”. The statement also notes that OpenAI is defending lawsuits over alleged plagiarism, copyright infringement and trademark dilution.
Tao added no opinion. His only words on the post are an editorial note marking it as a guest post and disclosing that the text was converted from another file format using AI.
What OpenAI released, and what came off
OpenAI published 722 preprints in a GitHub repository on 6 October, covering claimed progress on 372 mathematics problems across geometry, computer science, algebra and other areas, generated by an internal model it has not released.
A day later it withdrew three manuscripts over a sign error, Retraction Watch reported. The error invalidated an argument in one manuscript and the construction that two dependent papers relied on. OpenAI also revised 14 other manuscripts with proof repairs, corrected statements, clearer hypotheses and dependencies, and one correction to an obsolete citation. The catalogue now stands at 719.
One error, three papers
The shape of the failure is the part worth noticing. A single sign error did not stay inside one document. It propagated to two others that had built on the construction, which is what happens when a large body of work is produced at once by a system that reuses its own intermediate results.
Human mathematics has the same dependency structure. What it does not usually have is 722 documents arriving on the same day, with no reviewer having read any of them.
Not everyone agrees
The Advisory Group on Mathematics and Artificial Intelligence has taken a more measured line, treating the release as a first step whose mathematical content is only beginning to be understood. Cornell mathematician Alex Townsend said the errors were unsurprising and that he suspects more will be found, and argued that OpenAI should have announced the formally verified manuscripts first.
That is a procedural criticism rather than a boycott, and it points at the same problem: the release was ordered by quantity, not by confidence.
What to watch
The withdrawal rate. OpenAI has given the field a public, numbered corpus and a public correction log, which means the proportion of the 719 that survives review is measurable in a way almost no claim in this field is. Three down and fourteen revised on day one is the first data point, not the last.
