Article

Fields Medalists: severe misalignment of AI in mathematics

Draft. In the feed as a declaration — not a cyber incident.

On 11 September 2026, twenty-five Fields Medalists released A Severe Misalignment of AI in Mathematics (Tao’s blog; mathandai.org). The title word collides with this site’s name. Treat it carefully: the statement is about proxy and incentive misalignment in research math — labs racing famous problems as benchmarks against conceptual understanding, transmission, and attribution — not rogue models or containment folklore.

The declaration (short)

Company goals and community goals are “severely misaligned,” the signatories write. Famous problems have been lighthouses for methods; solving is a tool and proxy for insight, not the goal itself. Mass true/false output at lab tempo risks rush announcements, weak writeups, attribution fights, and a broken human transmission chain. AI could still help; outcomes depend on “the decisions of the humans in control.” Initial list is all Fields Medalists (Tao, Scholze, Viazovska, Bhargava, Avila, and twenty others); further signatures invited Leiden-style.

Discourse

The news is not only the letter. The argument is already splitting in public — accumulate takes rather than flatten them.

Update (Noah Smith, May 2026 — earlier round of the same fight): Economist @Noahpinion asked, verbatim: “What's the downside of AI discovering all the math that can possibly be discovered? Math doesn't exist as a jobs program for IMO gold medalists.” That frames the downside as employment nostalgia. The Fields letter’s downside is different: loss of the process that produces understanding and transferable tools — even if every theorem eventually gets a machine checkmark.

Update (Susan Zhang, 12 Sep 2026): In the declaration-adjacent thread, @suchenzang wrote that “the best maths people (not those award and fame strivers) seem to have a knack for shockingly prescient humor,” quoting Vince Vatter pointing at a 2019 McSweeney’s satire that fantasizes replacing “publish or perish” with “publish and perish” and reputation-as-arcane-rumor. Read as commentary: the prestige/benchmark culture the Fields letter attacks was already a joke among mathematicians — labs optimizing awards-shaped proxies are late to a critique the field told on itself.

Our lean (not a third “side”): the useful axis is proxy vs objective, not jobs vs automation. Noah’s question is fair if the only cost is displaced Olympiad careers; the declaration’s claim is that rushing problem-flags can destroy the conditions for insight. Zhang’s pointer is a reminder that prestige games inside math predate the labs.

What this is NOT

Not an HF×OpenAI-shaped incident. Not a containment rumor. Not a verdict on any single AI proof. Leiden remains related prior art, not a substitute.

Sources

• Tao — A Severe Misalignment of AI in Mathematics
• mathandai.org
• Susan Zhang (@suchenzang) — Sep 12, 2026
• Vince Vatter quote → McSweeney’s “An Open Letter to the Mathematical Community” (2019)
• Noah Smith (@Noahpinion) — May 21, 2026
• TechCrunch
• Leiden Declaration