OpenAI releases over 100 AI-generated solutions to unsolved math problems

By Splitfeed EditorialLast updated Oct 6, 2026, 23:00 UTC

On October 6, 2026, OpenAI published results from an internal frontier model on open problems in mathematics, releasing Lean proof formalizations and supporting research details publicly on GitHub. The release contained more than 100 solutions to previously unsolved problems. This followed an earlier, smaller release of solutions to 10 open math problems, also posted to GitHub, which had already generated significant reaction in the mathematics community before the larger batch arrived.

The scale of what OpenAI is releasing matters for reasons that go beyond any individual proof. Unsolved mathematics problems are not just puzzles sitting in a drawer; many represent decades of accumulated effort by working researchers who built careers around them. When an AI system produces solutions in volume and a company posts them publicly in rapid succession, it disrupts the normal rhythms of mathematical credit, collaboration and priority. The core dispute is not simply whether the proofs are correct but who controls the pace and manner of disclosure, and what obligations, if any, a private company owes to a scientific community whose foundational work it is drawing on.

No specific published reactions from the Pro-AI, Anti-AI or Middle Ground camps have been collected yet, so what follows reflects the positions each would typically take on a story like this. A Pro-AI camp would likely celebrate the releases as evidence that AI systems are reaching genuine frontier capability in formal reasoning, and point to the public GitHub posting as a model of open scientific contribution. An Anti-AI camp would probably amplify the concerns already voiced to WIRED, where one mathematician described a perception of "mobster behavior" from leading AI companies, framing the releases as a form of institutional pressure that crowds out human researchers and treats a shared scientific commons as a competitive resource. A Middle Ground camp would likely call for negotiated norms around disclosure, asking whether AI labs should coordinate with mathematical institutions before releasing results at this volume and speed.

The argument is unlikely to settle until mathematicians and formal verification experts have had time to audit the Lean proofs in detail, and until OpenAI or other labs clarify what process, if any, governed the timing and sequencing of these releases. Statements from major mathematical societies or journals responding to the October 6 batch would be worth watching as a signal of whether the field moves toward accommodation or more organized resistance.