OpenAI has published the output of an internal frontier model that was pointed at roughly 4,000 open problems in mathematics. The result: 722 manuscripts, grouped into 372 result families, together with Lean proof formalizations and supporting artifacts, all released through a public GitHub repository.
The mathematics community's reaction has been anything but gratitude. The Association for Human Mathematics, in a statement dated October 7, described the release not as a demonstration of scholarship but as "a demonstration of power", and urged mathematicians to stop collaborating with OpenAI. Terence Tao amplified the call. Researchers told Nature they were "up in arms"; The Verge recorded one reaction as "pure insanity", noting that open problems of this kind underpin dissertations, grant proposals, job applications and years of planned research.
OpenAI has since withdrawn three of the manuscripts — a sign error invalidated one argument, and the constructions two dependent papers relied on — and has revised others. That is an acknowledgment that volume is not verification. Supporters counter that formalized, machine-checkable proofs are a step change for research, whatever the social disruption.
What is undisputed: the bottleneck in machine-generated mathematics is no longer producing results, but humanly checking, crediting and absorbing them. That problem now has a 722-paper backlog.
