OpenAI AI proof of Yang-Mills existence verified in Lean
A committee led by physicist David Gross confirmed an AI-generated proof for a Millennium Prize Problem was formally verified, though the proof's meaning remains under review.

The short version
- The proof addresses the 'existence' half of the Yang-Mills Millennium Prize Problem, which carries a $1 million prize.
- A committee led by David Gross verified the proof in Lean, a formal mathematical language used to ensure logical consistency.
- The committee stated they do not yet fully understand the proof and that the 'mass gap' portion of the problem remains unsolved.
- Critics have raised concerns about the private 'shepherding' process and whether the verified statement matches the official problem requirements.
- A parallel effort using Anthropic's Claude is currently attempting to solve the three-dimensional version of the same problem.
Why it matters
This event serves as a primary test case for how the scientific community will validate AI-generated mathematics. It highlights the distinction between formal verification (checking logic) and human interpretation (confirming the result solves the intended problem).
The Yang-Mills Milestone
On November 9 and 10, 2026, a committee appointed by OpenAI announced that an AI-generated proof for the existence part of the Yang-Mills problem had been formally verified. This problem is one of the seven Millennium Prize Problems established by the Clay Mathematics Institute, which offers a $1 million prize for its solution.
The Yang-Mills problem is divided into two distinct parts. The first part requires showing that a rigorous construction of the Standard Model's underlying theory is possible in four dimensions. The second part requires showing that a "mass gap" emerges from its equations. The current AI result only addresses the first half of this requirement.
OpenAI reported that its internal model reached this result on September 9, 2026. The company kept the result private while it was being encoded into Lean, a specialized programming language used for formalizing mathematical proofs to ensure every step follows logically from its assumptions.
The Committee and Verification
To manage the release of the proof, OpenAI established a group of "shepherds" led by physicist David Gross. This committee includes four other colleagues, including Michael R. Douglas of Harvard. The group was tasked with bringing the AI-generated argument into the formal mathematical literature.
The committee issued a statement confirming that the result demonstrates the existence of pure Yang-Mills theory according to the standard problem statements. However, they were notably cautious in their wording, stating, "We do not claim to fully understand the proof yet." They used OpenAI’s Astra model to help them parse the complex Lean code.
As reported by CASRAI, the verification in Lean confirms that a formal statement follows from its formal assumptions. This is a technical check of logic, but it is different from a human understanding of the proof. Experts must still determine if the formal statement actually captures what the Clay Mathematics Institute intended for the original problem.
Process and Transparency Debates
The method of releasing the proof has sparked significant debate within the mathematical community. Some researchers view the "shepherding" arrangement as a form of gatekeeping. NYU mathematician Scott Armstrong noted that while he welcomed the Lean release, the review process could be conducted more efficiently in an open environment.
Critics like physicist Peter Woit have raised concerns about the substance of the proof. He argued that the existence verified might not meet the specific standards required by the Millennium problem, citing descriptions of "weak existence." He also noted that the committee has said nothing about the mass gap or other Standard Model results.
Transparency issues have also emerged regarding the privacy of the documentation. Woit noted that the committee kept a 1,147-page explanatory document private. Within hours of the Lean release, other groups began using AI agents to create public, readable versions of the code to bypass this private review.
Parallel Efforts and Next Steps
While OpenAI focuses on the four-dimensional problem, other researchers are pursuing related goals. Levent Alpöge from Anthropic and Ethan Sussman from Northwestern have been using Claude-based agents to work on a three-dimensional version of the existence problem. However, the Clay requirements specifically call for four dimensions, meaning their current progress would not qualify for the prize.
The release of the Lean code allows outside experts to begin scrutinizing the proof independently. This may happen through preprints and public repositories before any official journal decision is made. This process provides a live example for institutions drafting new policies on how to handle AI-generated research and peer review.
Several questions remain open as of November 12, 2026. The community must still decide if the Lean statement matches the Millennium problem's requirements and if the proof can be rewritten for human consumption. Additionally, the harder "mass gap" portion of the problem remains entirely untouched by the current AI result.
What to watch
- Whether outside experts confirm the Lean statement matches the official Millennium problem requirements.
- If the proof can be successfully rewritten into a format that humans can follow and understand.
- Progress on the 'mass gap' portion of the Yang-Mills problem.
Try it
Look up the Clay Mathematics Institute's Millennium Prize Problems to see the full list of unsolved problems and their $1 million rewards.
Sources and links
- ReportingCASRAI: the article we sourced
Liked it? Say why
Add a line for other readers.
No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.
Listen to the AI Wire headlines
Listen to the AI WireListen to our daily AI shows every morning: on your commute, in the shower, wherever you are
Listen to the daily showsFirst reported by CASRAI · Read the original. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.
