OpenAI forms math advisory group as its AI resolves more than 100 open problems

Artificial intelligence has breached one of the most guarded bastions of human intellect, sparking an immediate organizational defense from the industry's largest player. On Monday, 21 September 2026, OpenAI announced the formation of an independent advisory group hosted at the Institute for Advanced Study in Princeton, New Jersey. The move follows a blistering period of development in which an unreleased internal model, trained over a remarkably short span starting on 28 August 2026, resolved more than 100 long-standing open problems across almost every major discipline of mathematics. Among the casualties of this computational sweep is the Navier-Stokes Millennium Prize problem, a puzzle that has resisted human solution for decades.

The speed of the model's evolution caught internal researchers off guard. According to Eu.36kr, the system went from its initial training date to the public disclosure on 21 September 2026 in just 24 days. Behind this acceleration was a large-scale collaborative framework leveraging more than 10,000 concurrent AI agents. These agents autonomously accessed cached internet caches, executed code, and communicated across distinct solution pathways. Following initial success with Euler equations, researchers concentrated resources on the Navier-Stokes equations, utilizing Codex to aggregate intermediate insights across teams. A specialized model variant, GPT-6 Astra, completed the formalization and verification work in the Lean programming language.

Media additions

The disclosures have severely strained relations between commercial AI laboratories and elite academics. Just days after the Navier-Stokes breakthrough, recipients of the Fields Medal—the highest honor in mathematics—published a scathing open letter titled "The Severe Misalignment of AI in Mathematics." Reported by India Today and Moneycontrol, the signatories argued that treating centuries of accumulated human problems as mere leaderboard benchmarks threatens the foundational culture of the discipline. The letter warned that rushing out solutions without proper peer review, context, or attribution risks burying the mathematical community beneath a wave of unverified outputs.

The friction surrounding the breakthroughs extends beyond philosophical disagreements over publishing standards. As detailed by Unite.AI and t2ONLINE, questions of research priority have emerged. New York University researcher Tristan Buckmaster stated that he and Anthropic researcher Levent Alpöge had been working on related problems using OpenAI's Codex tool when information regarding their progress leaked to OpenAI. While OpenAI maintained it had not viewed their unpublished work directly, the company acknowledged it could not completely rule out de-identified product feedback contributing to model adjustments.

To navigate this institutional crisis, the newly established Advisory Group on Mathematics and Artificial Intelligence brings together nine distinguished academics. The roster includes Fields Medalists Timothy Gowers and Martin Hairer, alongside theoretical physics luminary Edward Witten, alongside Nikhil Srivastava, Ulrike Tillmann, Ravi Vakil, Melanie Matchett Wood, François Charles, and Camillo De Lellis. Notably, De Lellis is the only member from the panel who also signed the initial protest letter by the Fields Medalists, as reported by TechCrunch.

The structure of the advisory panel grants its members substantial independence. Operating entirely without compensation from OpenAI, the participants retain the right to offer unsolicited guidance, issue public criticisms, and dictate their own membership changes. However, strict institutional limits remain in place. As noted in announcements from both OpenAI and the Institute for Advanced Study, the panel holds no decision-making power over corporate operations and is explicitly barred from advising the company on how to pace its internal mathematical research.

The broader implications of these developments touch on ongoing concerns across the technology sector regarding autonomous capabilities, alignment, and safety. With university departments and evaluation committees now confronting the reality of automated theorem proving, mathematicians face a fundamental paradigm shift.