The Washington Post reported on 2026-08-19 that OpenAI convened a closed-door summit of about 40 top mathematicians around a single question: "When AI can do mathematical research, what is left for human mathematicians?" The meeting was moderated by OpenAI researcher Bubeck. Attendees included Fields Medalist Jacob Tsimerman, Harvard mathematician Melanie Matchett Wood, and German mathematician Andreas Thom. No consensus was reached, but the meeting triggered an internal 'value crisis' in the mathematical community.
1. Four Possible Futures: Bubeck's Open Question
Bubeck described four possible futures, without offering an answer:
1. Mathematics becomes like software engineering: AI helps hundreds of people collaborate on a single problem; 2. Mathematics becomes like physics: massive compute and AI models replace particle accelerators, directly 'computing' new theorems; 3. Mathematics becomes museum curation: AI produces, humans select and interpret; 4. Mathematicians collectively pivot to AI safety.
Bubeck's own remark was notably clear-eyed: "We must put people, put mathematicians, first. Mathematics only makes sense when mathematicians learn something from it." Yet he admitted the 40 participants reached no consensus.
2. AI Can Prove Theorems, But Can't Explain How
Harvard's Melanie Matchett Wood wrote a human-version proof of OpenAI's unit distance conjecture result. Her observation was sharp: top AI models share a common flaw in explanation — "they talk at length about the easy parts and gloss over the hard parts."
She further noted that top AI models still cannot identify the genuinely difficult part of an argument and explain it clearly. The implication is severe: the core value of mathematics is *understanding results, not just proving them* — a proof that is not understood does not become part of the literature.
Andreas Thom offered a more precise distinction. Among OpenAI's 10 published results was a new construction about non-sofic groups, filling a gap between Thom's and Gábor Kun's papers from 2016 and 2019. The two subsequently published follow-up work simplifying and extending the AI's finding. Thom said: "AI solves problems in very clever, substantial ways. But the new concepts emerged only after humans participated in the proof process — that is not yet something AI does on its own."
3. Tsimerman Chooses AI Safety: A Symbolic Event
Jacob Tsimerman, the 2022 Fields Medalist, has not yet formally joined OpenAI but has chosen to work on AI safety research — itself a microcosm of mathematicians' situation. A Fields Medalist choosing safety research at an AI company embodies the community's value crisis: as AI proves more theorems, mathematicians question whether their own work remains irreplaceable, and some turn to the more urgent front of ensuring AI is not misused.
4. Context: The Math × AI Mainline of Late August
- 8/23 Terence Tao 'digesting' AI proofs + Palomar registry: a 90,000-line Lean proof of the Sunflower conjecture compressed to 15,000 lines;
- 8/23 OpenAI Astra completed 10 open mathematics problems in Lean 4 on 8/1, security-locked on 8/7 as 'critical';
- 8/22 Axiom Math (founded by a 25-year-old woman from Guangzhou) completed Lean 4 formal verification of a prime gap theorem;
- 8/23 Talagrand's convexity conjecture: a 31-year-old open case resolved via human-AI collaboration;
- 8/24 the 40-mathematician closed-door summit (this post).
- If AI can mass-produce theorem proofs, can 'number of theorems proved' survive as an evaluation metric?
- If AI can run counterexample searches in hours, does 'finding a counterexample' still count as original contribution?
- If AI can rapidly generate formalized proofs, is formalization skill still a core competency for mathematicians?
- Don't focus only on "AI accelerating proofs"; also ask "what value remains for mathematicians";
- Don't focus only on tools; also on institutions — how Fields Medals, ICM invited lectures, and top journals should respond to the AI-era value crisis;
- China's strength is 'many people + high problem volume,' but its weakness is relatively fewer original concepts. In the AI era this is a dangerous combination: AI will rapidly erase the volume advantage, but cannot easily replace the centrality of original concepts.
- Washington Post, 2026-08-19: "Mathematicians ask what's left for humans when AI can do math research"
- 36Kr, 2026-08-24: "The End of Mathematics: 40 Top Mathematicians Gather at OpenAI's Secret Meeting"
- Bubeck's personal homepage; Melanie Matchett Wood's personal homepage
- Original papers by Andreas Thom and Gábor Kun (2016, 2019)
The mathematical community's stance toward AI is shifting from an 'instrumentalist' view to a 'value' view: no longer just "what theorems can AI help me prove," but "what remains for mathematicians when AI takes over proving?" The summit's landmark significance: this question was, for the first time, discussed jointly by a leading AI lab and top mathematicians.
5. Implications for Math Education and Research Evaluation
These questions have no answers yet, but they are now officially on the table.
6. Lessons for the Chinese Mathematical Community
China's mathematical community has invested in AI4S (at the Aug 24 China AI for Science conference, multiple academicians flagged 'verification as the core bottleneck') and math-AI tooling (Tao's Palomar relay, Lean 4 formalization). But the summit's deeper message: