Gross committee affirms AI-generated Yang-Mills existence proof
On 9-10 November 2026 the Yang-Mills Working Group, a five-member committee chaired by Nobel laureate David Gross and appointed by OpenAI, said that the machine-generated proof of the existence of pure quantum Yang-Mills theory checks out in the Lean proof assistant. The group released two GitHub repositories: OpenAI's Lean formalisation and its own validation assessment, which states that the result proves existence according to the standard problem statements in physics and mathematics. The committee stressed that the mass gap half of the Clay Millennium Prize Problem remains unproved and that its finding is not a prize determination. The accompanying 1,147-page written proof had not been released.
Why It Mattered
This is the first time an AI-generated result on a problem of this standing has been put through an organised external review and survived it. The underlying claim was announced earlier under a shepherding arrangement precisely because the mathematics and physics communities had no procedure for adjudicating machine output at this scale; the November statement is the procedure's first verdict. Three features make it a reference point. First, the division of labour: machine-checkable Lean code carried the weight of verification, while human experts judged whether what was formalised answered the question actually asked - a split between correctness and significance that is likely to become the standard template for evaluating AI contributions to mathematics. Second, the limits were stated plainly: existence, not the mass gap, and no claim on the Clay prize, which separates a genuine result from the flood of weakly vetted machine output that had already strained the field's review capacity in 2026. Third, the written proof remained unpublished, leaving the human-readable mathematics - the part that produces understanding rather than assurance - outstanding, and prompting open disagreement among mathematicians about whether Lean verification is sufficient grounds for belief. A history of AI written in 2030 would cite this as the moment when machine-produced mathematics acquired an institutional acceptance process, and as the beginning of a durable argument about what verification is for.
Who Built It
OpenAI; Yang-Mills Working Group chaired by David Gross
Applications
- Mathematics
- Theoretical Physics
- Formal Verification