dotsfeed
← News

OpenAI says its models produced three proofs in math, CS, and physics, and hands them to outside shepherds

Verified· Oct 27, 2026Published Oct 27, 2026

OpenAI says internal models produced three research results, a math proof, a computer science disproof, and a physics existence proof, and outside expert groups will shepherd their verification.

What happened

On October 26, 2026, OpenAI said its internal models produced three research results: a proof of the local smoothing conjecture in every dimension (which also resolves the Kakeya conjecture and Stein's restriction conjecture in all dimensions; AI output obtained Oct 18, formalized in Lean Oct 19); a disproof of the Exponential Time Hypothesis (natural-language disproof Oct 16, Lean formalization Oct 17; verified in Lean); and a mathematically rigorous existence proof of certain interacting quantum field theories in four dimensions, including quantum chromodynamics and a unified completion of the Standard Model, including pure Yang-Mills existence (natural-language proofs Sep 9 and Sep 18, Yang-Mills Lean formalization finished Oct 16).

OpenAI is experimenting with a new release approach: external expert 'shepherd' groups organize verification and understanding. The math output is released directly (AI writeup plus Lean proof in OpenAI's public GitHub repository); the computer science group (led by Russell Impagliazzo and Ramamohan Paturi of UC San Diego and Ryan Williams of MIT) will publish the AI output while collaborating externally, with all raw data public and OpenAI funding a conference; the physics group (led by David Gross, with Michael Douglas, Arthur Jaffe, Igor Klebanov, and Nikita Nekrasov) will hold off releasing AI artifacts and write the result up itself.

Collaborations operate under four principles: transparency (no NDAs), no academic credit sought by OpenAI for the discoveries, recognition for the groups' understanding and writeups, and independence (voluntary, unpaid, members' conclusions their own).

Why it matters

OpenAI frames the AI manuscripts as raw research data that must still be converted into scientific understanding, and says it does not yet know which of the three release approaches will prove most useful, inviting feedback. The results are not yet independently verified by the wider community; the shepherd model is OpenAI's attempt to handle verification, credit, and access for AI-generated breakthroughs.

Verified October 26, 2026.

Sources

Get updates like this every morning

  1. ① Email
  2. ② Card on Stripe
  3. ③ 7 days free

Then $2/month · cancel anytime in one click