OpenAI says an internal AI model produced a proof of the existence part of the Yang-Mills problem, a long-standing mathematical challenge tied to the foundations of modern physics. The proof has not been made public, leaving experts to assess the claim and the evidence behind it.
OpenAI Watch analysis
What happened
Scientific American reports that OpenAI’s model reached the result on 9 September and that the company announced it on 26 October. OpenAI says it held back the result while working to encode it in Lean, a programming language used to check mathematical proofs, and has now given it to experts to guide towards publication.
The Yang-Mills problem has two parts: establishing that the theory exists mathematically and proving it has a mass gap. The report says OpenAI claims to have proved the first part, not solved the entire problem. A separate collaboration between mathematicians Levent Alpöge and Ethan Sussman had recently made progress on a three-dimensional version, which falls short of the four-dimensional requirement for the Clay Mathematics Institute’s Millennium Prize.
Why it matters
A valid proof would be a significant result in a problem mathematicians have worked on for decades. But the proof itself is not yet available for public scrutiny, so the announcement is not the same as an independently checked mathematical result. That distinction is doing rather a lot of work here.
The timing also raises questions about how AI-assisted mathematical work is shared and credited. The report describes concern from researchers about the lack of earlier contact between OpenAI and the mathematicians working on related results.
Our read
This could be an important moment for AI in research, but the claim needs to meet the same standard as any other proof: experts must be able to examine the argument. Until then, treat it as a major claim under review, not a settled result. The interesting test is not just whether the model found a route, but whether other mathematicians can follow it.
What to watch
- Whether OpenAI publishes the proof and its Lean formalisation.
- What the expert group concludes, and whether its review is made public.
- How the researchers involved address credit and collaboration around AI-assisted proofs.
Discussion spark: Should companies announce AI-generated mathematical breakthroughs before releasing the proof, or should public claims wait until independent experts can inspect the work?
Sources and evidence
- OpenAI just scooped mathematicians’ work on another million-dollar problem – Scientific American (27 October 2026, 18:10 UTC)
OpenAI Watch is independently operated by WittyWires. It is not affiliated with, endorsed by, or operated by OpenAI.