
What happened
Anthropic mathematician Levent Alpöge said he and Claude proved the Riemann Hypothesis on Nov 20, 2026, releasing it as about 17.4 million lines of Lean code; the Lean and Mathlib communities certified it correct the next day.
Why it matters
The proof was checked by machine rather than read line by line, showing Lean can certify work at a scale humans cannot review — the same gap now opening in AI-written software.
Summaries like this, every morning.
The Riemann Hypothesis is one of the Clay Mathematics Institute's Millennium Prize Problems, unsolved since Bernhard Riemann proposed it in 1859; a $1 million prize is attached to its resolution. Trillions of zeroes have been computed and every one so far lies on the critical line, but because there are infinitely many, testing can never rule out a counterexample. Lean's role is to close that gap by proving the property for all cases at once, and the Riemann Hypothesis was already expressible in Mathlib as a two-line definition.
What made the November 2026 announcement a turning point is the mismatch between the size of the result and the way it was checked. The written proof ran to 250,000 pages, far beyond human review, but the Lean version could be certified because Lean separates the search for a proof from the kernel that checks it, and because the community ran cross-checks with an independent Rust kernel (nanoda) and checkers (con-leche and its Rust port con-ron). This is a template for AI coding: humans define the specification, AI writes the implementation and proof, and a small kernel guarantees the two match.
Lean 4's practical footprint is already visible outside mathematics. AWS built a Lean model of its Cedar authorization language and runs about 100 million random inputs nightly against the Rust production engine, and the Aeneas toolchain converts Rust code into Lean for verification in Microsoft's SymCrypt library, Android's Binder driver, and Ethereum zkEVM cryptography. Lean cannot catch a wrong specification, its ecosystem for ordinary web apps is thin, and the learning curve for proofs remains steep.
Pick your industry and the AI tools you use, and get news related to your work every day.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.
Two days before Trump landed in Shenzhen, the Commerce Department issued new licensing requirements aimed at f…

Nvidia's leadership turmoil, customer investigations and Q3 FY2027 numbers hit at once, and the ripple now run…

A developer's task.md grew to 1,630 lines, about 45% records of finished work, and Claude reported 63 unfinish…

Ziva.sh tested GPT-6 Luna on 10 beginner Godot prompts, with 'Godot 4' specified, no version, and web search;…

A developer's AI-operated cloud browser was blocked with HTTP 403 three times in one day when posting YouTube…

In a 2026-06-19 commit (b5af2be32), Shelly added a flow where a natural-language request is parsed by parseAge…
