In September, OpenAI said an unreleased frontier model had resolved more than 100 long-standing open problems across much of mathematics. As anticipation built, the newly formed Advisory Group on Mathematics and Artificial Intelligence urged labs to publish promptly through established academic channels, disclose models, prompts and compute, and avoid turning results into marketing.¹

On Tuesday, OpenAI released 722 manuscripts, arranged into 372 families of related findings. The company supplied paper-revision and citation protocols through a GitHub repository, along with summaries of reasoning, compute estimates and statistics on attempted problems. It said the average result consumed the equivalent of three hours of ChatGPT Pro thinking.¹

OpenAI executives framed the release as part of a larger scientific project. Greg Brockman called it a move “towards acceleration of scientific discovery and improving quality of life for everyone.”Greg Brockman (gdb)² The volume alone impressed observers: hundreds of proofs generated by an unreleased model suggest mathematics could follow programming into a world where machines are no longer a novelty but a working force in specialist fields.³

But enthusiasm has been matched by alarm. Researchers are still asking whether the results are genuinely original or rely heavily on prior human input, and whether a torrent of machine-produced papers can be meaningfully assessed. Stephen Wolfram captured the quality-versus-quantity objection: “You can make a trillion theorems easily. The problem is most of those theorems are not ones that anybody will care about.”³

The distrust is not merely philosophical. Earlier OpenAI announcements had already triggered allegations involving unpublished work, scooping and weak disclosure around training-data origins. Critics fear a wealthy AI lab can race through problems researchers have spent careers pursuing, without the academic norms that govern attribution and verification. One professor described the conduct as the “kind of things that mathematicians will generally not do.”⁴

Even the wider online response reflected that caution: Elon Musk amplified a post insisting, “No, AI didn’t just solve the Navier-Stokes equations.”Elon Musk (elonmusk)⁵ The new release may be an extraordinary technical milestone; whether it becomes a trusted contribution to mathematics remains a human question.