Terence Tao and 24 Fields Medalists Issue Historic Declaration Challenging OpenAI Over AI Math Claims

A high-stakes philosophical and scientific confrontation has erupted at the intersection of pure mathematics and Silicon Valley. Following aggressive corporate disclosures by OpenAI claiming autonomous breakthroughs on a Millennium Prize problem, twenty-five Fields Medal recipients—widely regarded as the Nobel laureates of the mathematical sciences—have taken the extraordinary step of issuing a joint public declaration: 'A Severe Misalignment of AI in Mathematics.'

Published on September 11, 2026, at mathandai.org, the open letter represents the most authoritative rebuke of commercial AI research culture to date. The initial signatories include UCLA professor Terence Tao, widely celebrated as the greatest living mathematician and a prominent pioneer of AI-assisted formal verification, alongside legendary luminaries spanning five decades of mathematical breakthroughs: 1978 winner Pierre Deligne, 2010 medalist Ngô Bảo Châu, and 2026 recipient Deng Yu. The declaration warns that commercial AI laboratories, driven by investor hype and marketing cycles, are pursuing benchmark victories that directly undermine the foundational integrity of scientific discovery.

The immediate catalyst for the confrontation was OpenAI's high-profile announcement that its frontier reasoning model, GPT-6 Astra, had autonomously formulated proofs targeting Millennium Prize problems—specifically conjectures surrounding the Navier-Stokes equations. While OpenAI framed the milestone as proof that artificial general intelligence is unlocking the holy grails of science, the academic mathematical community reacted with immediate skepticism.

Unlike traditional mathematical papers submitted with rigorous, line-by-line exposition and verifiable derivations, OpenAI's announcement was accompanied by promotional press statements and selective excerpts rather than an end-to-end formal proof. Crucially, the outputs lacked machine-checkable formal verification in interactive theorem provers such as Lean 4 or Coq—the exact gold standard required to prove that complex symbolic arguments are free of subtle semantic hallucinations.

For Terence Tao, who has spent the past two years championing the formalization of mathematics in Lean and deploying language models to assist with collaborative polymath projects, the issue is not hostility toward artificial intelligence. Rather, it is the weaponization of mathematics as a disposable corporate PR vehicle.

The declaration outlines four structural misalignments that threaten the continuity of mathematical research:

1. 1.The Destruction of Attribution Norms: Commercial models are trained on centuries of peer-reviewed mathematical scholarship without formal attribution, licensing, or consent. When an AI generates a derivation, corporate press releases routinely claim sole discovery for the model, erasing the lineage of human mathematicians whose decades of foundational work made the breakthrough conceptually possible.

1. 1.The Inundation of Unverified Proof Slop: In mathematics, a proof that is 99% correct is completely incorrect. A subtle flaw in a single lemma can invalidate an entire 200-page treatise. Mathematicians warn that flooding preprint repositories like arXiv with plausible-sounding, multi-hundred-page AI-generated arguments forces human peer reviewers to spend months hunting down hallucinated steps, effectively paralyzing the academic review apparatus.

1. 1.Understanding Over Output: In the words of the declaration, mathematics is not a collection of boolean true/false answers; it is a discipline built on human comprehension, conceptual tools, and the transmission of insight between generations. An unverified or incomprehensible token stream that claims to solve a conjecture without conveying reusable geometric intuition provides zero enduring scientific value.

1. 1.Closed Models and Corporate Hegemony: The concentration of compute-intensive frontier models within a handful of commercial firms threatens the open, decentralized ethos that has defined mathematical progress since antiquity. If scientific proofs can only be generated or audited using proprietary, closed-source APIs costing tens of thousands of dollars in compute, academic independence collapses.

The Fields Medalists are not calling for an AI moratorium, but for enforceable scientific standards. The declaration lays out clear protocols that commercial labs must adopt before claiming mathematical discoveries:

• Machine-Checkable Verification: Major mathematical claims must be accompanied by complete, compilable code in open formal verification systems like Lean 4, allowing the global mathematical community to programmatically verify validity without relying on corporate self-audits.
• Full Open-Access Preprints: Labs must release complete mathematical write-ups detailing definitions, intermediate lemmas, and counter-example bounds simultaneously with public announcements, rather than withholding derivations behind marketing embargoes.
• Attribution Heritage Mapping: AI outputs must cite the specific historical papers, conceptual frameworks, and human theorems utilized by the model's reasoning trajectory.

The battle over mathematics carries urgent implications far beyond university lecture halls. Mathematics represents the purest form of deterministic logic; if frontier AI models cannot be trusted to maintain truth and attribution in a domain governed by absolute formal rules, the premise of deploying autonomous agents into messy, high-stakes real-world environments like clinical medicine, legal contracts, and financial infrastructure becomes deeply suspect.

By drawing a line in the sand, Terence Tao and his colleagues have sent a clear message to Silicon Valley: true scientific discovery cannot be automated with marketing shortcuts. In the quest for frontier intelligence, rigor is non-negotiable.

• A Severe Misalignment of AI in Mathematics: The Declaration — Math & AI Coalition
• A Severe Misalignment of AI in Mathematics — Terence Tao's Blog (What's New)
• OpenAI's feud with mathematicians is only escalating — TechCrunch
• Can AI Solve the Hardest Problems in Mathematics? — The Economist