The physicists OpenAI asked to handle its claimed Yang–Mills result said they would publish OpenAI’s Lean code on Tuesday, 10 November. Their third statement, dated 9 November, names two GitHub repositories: the “Full Lean code submitted by OpenAI to the group on Oct 28, 2026” and the group’s own validation document.
The committee is chaired by Nobel laureate David Gross, with Michael Douglas, Arthur Jaffe, Igor Klebanov and Nikita Nekrasov, and its statements are hosted by the Kavli Institute for Theoretical Physics. When we checked GitHub at 20:19 UTC, 12:19 in California, neither repository was public yet.
Business Pill · The AI Audit Problem
A short explainer on the AI audit problem: when producing work becomes cheaper than checking it. It teaches the general idea only and does not discuss OpenAI, the committee or this proof.
The key insight: The Yang–Mills shepherding committee chaired by David Gross set 10 November to publish the Lean code OpenAI submitted on 28 October, with its own validation document. It describes the result as the existence of pure Yang–Mills theory and says confinement and the mass gap remain open.
The 9 November statement describes the validation document as one “which summarizes the results and our checks of this submission, specifically that it proves the existence of pure Yang-Mills theory according to the problem statements which are standard in physics and mathematics”.
The committee’s first statement says OpenAI asked Gross on 16 October “to assume custody of the proof, decide on the best way to release it to the community, and shepherd the release process”, and that Gross then chose the rest of the group. It adds: “We have no responsibility to OpenAI”.
The same statement says the new proof “appears to be an important step towards putting QFT on a rigorous mathematical footing”, and that before releasing it “we would like to be convinced that it is correct”. It credits earlier work, naming Kenneth Wilson’s 1974 lattice approach and Tadeusz Balaban’s papers from the 1980s. It also says: “But confinement and the mass gap remain very important open challenges”.
The second statement, of 3 November, said the group would “announce any release of LLM produced material a full day in advance”. The 9 November notice is that announcement.
Days between the steps OpenAI and the Yang-Mills shepherding committee have dated, from the claimed natural-language proof on 9 September to the 10 November release date the committee set for the Lean code. Our arithmetic.
Our Numbers
OpenAI has said its models produced the natural-language proof of Yang–Mills existence on 9 September and finished the Lean formalization on 16 October, as we reported on 26 October. From 9 September to the 10 November release date is 62 days, by our arithmetic: 37 days to the Lean proof, 12 more until OpenAI submitted it to the group on 28 October, and 13 from submission to the planned release.
The release is the Lean code, not the full write-up. The first statement says the committee will provide its assessment “together with the full 1147-page LLM proofs” once it is “satisfied with the quality and correctness of the proofs”. It gives no date for that.
Why It Matters for the AI Business
OpenAI said pure Yang–Mills existence is part of a Clay Mathematics Institute Millennium Prize problem. Terence Tao told The Observer that OpenAI’s push on such a problem “became code red priority number one” after a rumour that Anthropic was close to one; that is his reading of events. The rival lab’s model is in a similar story this week: PSI says an Anthropic model proved an 18-year quantum conjecture.
As we read it, both labs now present research results as evidence of what their models can do, and Lean is the part that lets outsiders check the work without trusting the lab. A separate public GitHub project, created on 10 November, aims to turn the Lean formalisation into a human-readable proof; its page says the Lean source “has not been imported or checked here”.
The Structural Read
As we read the statements, the committee is treating the Lean code as the first checkable artefact: it says it wanted to confirm that the definitions and problem statements are the standard ones before release, and the 1,147-page natural-language proofs come later.
As we read the timeline, the claim has spent 62 days with OpenAI and then with outside reviewers before any code goes public, by our arithmetic, which is the window in which a lab’s research result can be cited before anyone else can check it.
Yang–Mills shepherding committee, first statement, October 2026
“This is an experiment in how the academic community can deal with the rapid development of new knowledge by AI”
Three Implications
FOR RESEARCHERS The committee says it will publish the Lean code and a validation document now, and the full 1147-page proofs once it is satisfied with their quality and correctness.
FOR OPENAI The committee says it has no responsibility to OpenAI and works on behalf of the academic community.
FOR THE MODEL RACE Terence Tao described OpenAI’s Millennium Prize push as a race with Anthropic; that is his reading, as reported in our earlier piece.
The framework lists: “The Verifier Moat — built by RLVR on domain-specific reward signals, building into a new asset class for reasoning-heavy verticals.”
As we read it, a Lean proof is a verifier of exactly that kind: a check a machine can run. The framework does not discuss Yang–Mills or this release.
What Is Not Established
We have seen no peer-reviewed publication of the proof, and we found no statement on it from the Clay Mathematics Institute. The committee’s own wording limits the claim to existence: it says confinement and the mass gap remain open. OpenAI and the Advisory Group on Mathematics and Artificial Intelligence both describe existence as part of a Millennium Prize problem, not the whole of it. The repositories were not public at the time of our check, so we have not read the Lean code or the assessment. That advisory group, as we reported, had called on OpenAI to make its outputs public “without delay”.
This Week In The AI Supercycle
The Great Convergence of Capital, Power, and Geopolitics
This week’s issue: for two years the AI build-out was priced as a supply story, and last week as a credit story. This week it was priced as a permission story, and the answer arrived as a cost, not a ban.
The Gross-led committee that holds OpenAI’s claimed Yang–Mills existence proof said it would publish OpenAI’s Lean code and its validation document on 10 November, 62 days after OpenAI says its model produced the proof. The committee says the result covers existence; it says confinement and the mass gap remain open, and the full 1,147-page proofs will follow later.
A note on sourcing. Quotations are from the three statements of the Yang-Mills shepherding committee published by the Kavli Institute for Theoretical Physics. OpenAI’s dates are from its 26 October post, as covered in our earlier piece. The GitHub check is our own, at 20:19 UTC on 10 November. Day counts are our arithmetic. OpenAI and the committee were not contacted for this piece.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
