Levent Alpöge claims, with Claude, a proof of the Riemann Hypothesis; experts have not checked it
He says Lean checks it with no assumptions. Tao calls it enormous; a separate human-written proof of the GRH surfaced the same day.
On 20 November 2026 Levent Alpöge, a mathematician at Anthropic, published a short note saying that he has, “working with Claude”, found a proof of the Riemann Hypothesis, the best-known open problem in number theory. He says the method extends to the Generalized Riemann Hypothesis (GRH), a version covering a larger family of functions, and he expects it to reach the Extended version too.
The note says he and Claude “formalized the arguments without assumptions” in Lean, a language in which a computer checks every step. It says the formalization is finished. Alpöge chose to release only the Lean files, split into 14 archive parts, because a full written version would be too long to read; he says he plans to write a proper one.
Terence Tao wrote the same day that Alpöge, “using an internal AI model”, produced an “(enormous) Lean-only proof”. He expects checking to take some time and points to a Lean Zulip thread, where moderators are keeping the discussion to the Lean side. The note names no model version.
The same day Ravi Vakil announced a separate draft proof of the GRH by Alessandro Masullo, written without AI. Vakil says Brian Conrad has been checking it for months and that it is “so far looking good”.
What it means
If the Lean files hold and their statement matches the standard one, an AI-assisted proof has settled the Riemann Hypothesis. Until mathematicians confirm both, the problem counts as open.
Caveats
- None of the sources says anyone has compared the Lean statement with the standard statement of the hypothesis. A proof of a misstated claim would pass Lean and prove nothing.
- No expert has reported verifying the proof. Early comments under Tao’s post disagree: one commenter’s AI audit was unimpressed, another found it promising, and none says it built the archive.
- The note gives no line count for the formalization, and no human-readable argument exists yet.
- Alpöge wrote the note as an announcement, and his employer builds the AI he credits.
- Masullo’s draft is a separate claim, still under review.