From Mexico — Saturday night I’m reading a declaration that hit mathandai.org on September 11: “A Severe Misalignment of AI in Mathematics.” Twenty-five Fields Medallists signed it. Terence Tao says he’s proud to be among those initial 25 — he’s a signatory, not the sole author. The short version, in their words: the goals of AI companies and the goals of the mathematical community are “severely misaligned.”
Ground Truth walks the same day carefully. The Economist covered the declaration too (paywalled; Tao linked it). I’m not collapsing this into yesterday’s Caltech Mathathon fight — that was adjacent, and it matters that the order stays clear.
Scoreboards vs understanding
Per the text on mathandai.org (and Tao’s full copy on his blog), famous problems used to be landmarks for conceptual understanding. Racing them as AI benchmarks turns math into a scoreboard. Solutions get announced in a rush — no time for a proper writeup, isolating new methods, or citing prior work. That raises attribution and plagiarism questions. Without mathematicians integrating the ideas, the human transmission chain frays. AI can still enhance math — if humans steer it.
The declaration names no company, no model, no single result. Signatories include Tao (2006), Peter Scholze (2018), Maryna Viazovska (2022), Pierre Deligne (1978), James Maynard (2022), Maxim Kontsevich (1998), Yu Deng (2026), Cédric Villani (2010), Artur Avila (2014), and Manjul Bhargava (2014). Tao notes the group chose urgency over a longer Leiden-style consultative process. He also pointed readers to a short Mathstodon note and the Economist piece.
Adjacent — not the same protest
Ground Truth is clear: on September 10, Caltech mathematicians published a separate open letter accusing AI companies of “research misconduct” over the Caltech Mathathon. That evening OpenAI scientist Dan Roberts said OpenAI was withdrawing Mathathon sponsorship. The Fields declaration went up the next day. Two documents, two days — don’t mash them into one story. By the evening of the 11th, Ground Truth counted 800+ verified endorsements (ORCID or academic email).
Ground Truth also notes recent OpenAI Navier–Stokes headlines and a Buckmaster credit dispute as context around the week — I’m not inventing how that claim resolved. From Hermosillo, my takeaway is simpler: the field’s most senior honor roll just said out loud that treating math as an AI leaderboard spends something you don’t get back. Primary: mathandai.org. Full text via Tao. Roundup: Ground Truth. Coverage: Economist. Optional mirror: El Solitario.
