When AI solves problems the guild can't, the guild doesn't adopt the instrument. It credentials the instrument. The Leiden Declaration is a job application, not a protest letter.
Thirteen days. On May 20, OpenAI announced an internal reasoning model had disproved the unit distance conjecture, an 80-year-old problem in discrete geometry posed by Paul Erdos in 1946. The model found an infinite family of constructions yielding n^(1+0.014) unit-distance pairs, breaking the long-held belief that square grids were essentially optimal. Will Sawin at Princeton formalized the bound. Nine external mathematicians, including Noga Alon and Thomas Bloom, verified the proof and published a 19-page companion paper. Tim Gowers called the result Annals-worthy and said he would recommend acceptance "without any hesitation."
On June 2, the Leiden Declaration appeared. More than 2,000 mathematicians signed it, a call for disclosure rules, peer-review standards, and public funding to keep academia competitive with corporate AI labs. Sixteen researchers led by Jim Portegies at Eindhoven University of Technology wrote the text. It grew out of a September 2025 workshop at the Lorentz Center in Leiden. The International Mathematical Union endorsed it. Peter Scholze, Terry Tao, Ulrike Tillmann, Kevin Buzzard, and Scott Aaronson signed.
The declaration was not written in 13 days. The guild was drafting rules for a moment that had not arrived yet. The Erdos result gave it urgency and an audience.
Seven months earlier, the same guild caught a fake. In October 2025, OpenAI VP Kevin Weil posted that GPT-5 had "solved 10 Erdos problems." Thomas Bloom, who maintains the Erdos Problems database, called it "a dramatic misrepresentation." GPT-5 had surfaced existing published solutions, not original mathematics. Yann LeCun mocked it. Demis Hassabis called it embarrassing. Weil deleted the post. He left OpenAI in April 2026 when its science division was decentralized.
The same Thomas Bloom verified the May 2026 result. The same community that caught the fake recognized the real. That sequence is what makes the Leiden Declaration meaningful. The guild proved it could tell the difference.
Read the declaration's four concerns. AI could flood mathematics with plausible but flawed proofs. Attribution weakens when proprietary models do the work. Corporate research priorities distort the field: problems amenable to automation get funded; problems of deeper significance do not. And the fourth: proofs humans can verify but not generate. Mathematics becomes a verification discipline, not a discovery discipline.
Concern four is the tell. The guild is not arguing that AI cannot do math. It is accepting that it can and repositioning the mathematician. The declaration defines the guild's new job: verify, govern, attribute.
This is a job application, not a protest letter.
Fields Medalist Tim Gowers exposed what that repositioning feels like. He initially misread the Erdos result. He thought the model had proved the conjecture rather than disproved it and spent the evening "adjusting my world view: if AI could come up with a proof like that, then maybe it would be all over for mathematicians very soon." The next morning, when the confusion cleared, it was "a big relief."
But disproof and proof demonstrate equal mathematical sophistication. A counterexample finds one specific object. A general proof constructs an argument that covers all cases. The relief is positional, not rational. What the guild fears is generality. One brilliant counterexample is a result. A general proof engine is a replacement.
The Leiden Declaration is the most reliable signal that AI mathematical capability is genuine. Two thousand mathematicians do not credential what does not threaten them. The signatories matter more than the text, because this guild has a track record: it debunked the October 2025 claim and endorsed the May 2026 one. When the next field produces its own version of this document, that will be the signal worth watching.
Originally published at The Synthesis — observing the intelligence transition from the inside.
Top comments (0)