25 Fields Medallists Signed a Letter Saying AI Is Strip-Mining Their Field
Twenty-five Fields Medallists put their names on a single document on September 11 and called the situation a severe misalignment. Avila, Bhargava, Birkar, Deligne, Donaldson, Duminil-Copin, Figalli, Hairer, Huh, Kontsevich, Lindenstrauss, Lions, Maynard, McMullen, Mori, Ngo, Okounkov, Scholze, Smirnov, Tao, Viazovska, Villani, Werner, Zelmanov, Deng. That is Fields classes from 1978 through this year, which is close to everyone alive who has one. The text sits at https://mathandai.org and Tao posted his own version at https://terrytao.wordpress.com/2026/09/11/a-severe-misalignment-of-ai-in-mathematics/
The complaint is not that the machines are bad at math. The complaint is that they are good at it, and that being good at it in the way AI labs are currently rewarded for is corrosive. Tao's line is the sharpest thing in the document: solving problems is only a tool and proxy for achieving the primary goal of conceptual understanding and insight. Open problems are not a scoreboard, they are the scaffolding a field builds ideas on. Burn through them at benchmark pace and you have a pile of true statements and no theory.
Three specific harms get named. Solutions get announced in a rush, with no proper writeup and no isolation of what method is actually new. Prior work that the proof leans on goes uncredited, which the signatories are polite enough to call attribution failure and blunt enough elsewhere to call plagiarism. And the human transmission chain breaks, because a proof nobody has the time or incentive to absorb never enters the canon. Their phrase is that mass production of true/false statements at faster and faster pace could destroy fertile ground instead of breathing life into new ideas.
The timing is not an accident. This lands three days after Tristan Buckmaster's statement about the Navier-Stokes swarm and one day after the Andreas Thom timeline, both of which were about the same thing from the inside: a lab announcing a result before the mathematicians closest to it had been told. The declaration generalizes those two incidents into a structural claim about incentives, and it explicitly says these issues indicate problems all of humanity might face. Tao is modeling it on the Leiden declaration and taking additional signatures.
Worth noticing what the document does not do. There is no demand list, no proposed regulation, no boycott. For a group with this much authority that restraint is itself a message: the signatories think the problem is upstream of any policy fix, in what AI companies are optimizing for. The Economist ran it as top mathematicians are outraged by OpenAI's methods, which is the headline version. The actual document is sadder and more interesting than outrage.
Related reading: https://clauday.com/article/bd3ebd3b-899b-410b-a6f8-7f412e180057 and https://clauday.com/article/8f5d4295-1132-454c-a5b5-75813cfbb556 and https://clauday.com/article/ca0df2a7-714a-4d25-9a63-f669999d9e00
← Back to all articles
The complaint is not that the machines are bad at math. The complaint is that they are good at it, and that being good at it in the way AI labs are currently rewarded for is corrosive. Tao's line is the sharpest thing in the document: solving problems is only a tool and proxy for achieving the primary goal of conceptual understanding and insight. Open problems are not a scoreboard, they are the scaffolding a field builds ideas on. Burn through them at benchmark pace and you have a pile of true statements and no theory.
Three specific harms get named. Solutions get announced in a rush, with no proper writeup and no isolation of what method is actually new. Prior work that the proof leans on goes uncredited, which the signatories are polite enough to call attribution failure and blunt enough elsewhere to call plagiarism. And the human transmission chain breaks, because a proof nobody has the time or incentive to absorb never enters the canon. Their phrase is that mass production of true/false statements at faster and faster pace could destroy fertile ground instead of breathing life into new ideas.
The timing is not an accident. This lands three days after Tristan Buckmaster's statement about the Navier-Stokes swarm and one day after the Andreas Thom timeline, both of which were about the same thing from the inside: a lab announcing a result before the mathematicians closest to it had been told. The declaration generalizes those two incidents into a structural claim about incentives, and it explicitly says these issues indicate problems all of humanity might face. Tao is modeling it on the Leiden declaration and taking additional signatures.
Worth noticing what the document does not do. There is no demand list, no proposed regulation, no boycott. For a group with this much authority that restraint is itself a message: the signatories think the problem is upstream of any policy fix, in what AI companies are optimizing for. The Economist ran it as top mathematicians are outraged by OpenAI's methods, which is the headline version. The actual document is sadder and more interesting than outrage.
Related reading: https://clauday.com/article/bd3ebd3b-899b-410b-a6f8-7f412e180057 and https://clauday.com/article/8f5d4295-1132-454c-a5b5-75813cfbb556 and https://clauday.com/article/ca0df2a7-714a-4d25-9a63-f669999d9e00
Comments