OpenAI Claims AI Proof of Navier-Stokes Millennium Problem Amid Academic Controversy

Same model resolved 100+ long-standing open math problems beyond Navier-Stokes

OpenAI's new internal model — training started August 28 — has now resolved more than 100 long-standing open problems across most areas of mathematics, far beyond the Navier-Stokes Millennium Problem. OpenAI states the pace of mathematical progress has surprised its own internal mathematicians.

OpenAI forms independent Advisory Group on Mathematics and AI, hosted at IAS

Responding to community concerns, OpenAI partnered with mathematicians to establish an independent advisory group hosted at the Institute for Advanced Study. The group will advise on review, significance assessment, and dissemination of emerging mathematical results. Members are unpaid, can offer unsolicited public advice, and operate independently from OpenAI.

Advisory group includes 9 prominent mathematicians — including Fields Medalists Gowers, Hairer

The nine initial members are: François Charles (ENS-PSL), Camillo De Lellis (IAS/GSSI), Timothy Gowers (Collège de France/Cambridge), Martin Hairer (EPFL/Imperial College), Nikhil Srivastava (Berkeley/Simons Institute), Ulrike Tillmann (Oxford/INI), Ravi Vakil (Stanford), Edward Witten (IAS), and Melanie Matchett Wood (Harvard).

Open letter 'A Severe Misalignment of AI in Mathematics' prompted the advisory response

An open letter titled 'A Severe Misalignment of AI in Mathematics' — signed by mathematicians — raised concerns about AI companies using solved open problems as benchmarks, citing negative externalities on the math research community. OpenAI cited this letter as a key motivation for establishing the advisory group.

Lean 4 proof verified in 17 hours vs. ~132,800 person-hours historically — 4 orders of magnitude faster

OpenAI verified the 166-page Navier-Stokes proof in Lean 4 in just 17 hours. Prior to AI, research-level mathematics took an estimated ~40 person-hours per textbook page to formalize (per 2005 Barendregt-Wiedijk analysis); the equivalent human effort for OpenAI's 166-page paper would have been approximately 132,800 hours — a reduction of roughly four orders of magnitude.

Lean 4 speedup could enable machine-checkable proofs for security policies, smart contracts, and critical code

The four-orders-of-magnitude reduction in formal verification cost could extend beyond mathematics. The same AI-driven formal proof capability could verify security policy consistency, smart contract liability limits, and mission-critical algorithm correctness — domains where machine-checkable proofs are valuable but previously too expensive to generate at scale.

Proof shows NS equations exhibit finite-time singularities — negative answer to regularity

MIT Technology Review clarifies the specific mathematical result: OpenAI's proof demonstrates the full Navier-Stokes equations can break down — exhibiting finite-time singularities where solutions become infinite in finite time. This constitutes a negative answer to the Millennium Problem's global regularity question: smooth solutions do not always exist. (Source: MIT Technology Review, September 9, 2026)

OpenAI will not claim the $1M Clay Mathematics Institute prize

Despite the reported proof, OpenAI announced it does not plan to claim the million-dollar prize from the Clay Mathematics Institute for solving the Navier-Stokes Millennium Problem. No explanation for this decision was provided in public reporting. (Source: MIT Technology Review)

~10,000 agents trained via multi-agent RL over ~1 year drove the solution

OpenAI-linked sources (Ethan Knight, cited by AINews/Latent Space) described the Navier-Stokes effort as a collaboration of approximately 10,000 agents trained via multi-agent reinforcement learning over roughly one year. Agents used large amounts of unstructured parallel test-time compute and self-organized to divide work — a distributed theorem search paradigm rather than a single long-chain proof attempt. AINews notes the '88 hours' figure that circulated widely came from a satirical post and should not be treated as confirmed. (Source: AINews/Latent Space)

Buckmaster posted his own simplified NS blow-up proof on Mastodon before OpenAI

On the Monday before OpenAI's announcement, NYU's Tristan Buckmaster posted a proof on Mastodon showing a simplified version of the Navier-Stokes equations can also break down. He and Alpoge had worked on the full problem for nearly a year. OpenAI then announced a proof for the full NS equations using an unreleased model that 'dramatically outperforms' the already-powerful Astra. (Source: MIT Technology Review)

Frontier math breakthroughs may now require resources only two labs possess

MIT Technology Review noted that solving the most important unsolved mathematical problems now appears to require AI resources available only at a handful of frontier AI companies — compute budgets and model capabilities that dwarf what individual academics or university labs can access. This raises fundamental questions about how human mathematicians will contribute to major breakthroughs in the AI era. (Source: MIT Technology Review)