AI

Anthropic Mathematician Levent Alpöge Posts Claude-Assisted Riemann Hypothesis Proof, and Lean Groups Certify It

658 words, 3 min

anthropic mathematician levent alp ge posts claude assisted pixabay 1547018

Image: Pixapopz on Pixabay

Tech news by email, once a week

Unsubscribe any time. Privacy policy

Levent Alpöge, a mathematician at Anthropic, published a four-page note on November 20, 2026, saying he found a proof of the Riemann Hypothesis while working with Anthropic's Claude model. The Lean and Mathlib communities followed on November 21 with a joint statement certifying that the formal proof is correct, according to published reports.

What Alpöge released

The note is titled "Remark" and is hosted on Alpöge's personal site. In it he writes that he believes he has, working with Claude, discovered a proof of the Riemann Hypothesis. He says the method extends to the Generalized Riemann Hypothesis and expects to reach the Extended Riemann Hypothesis "with more time."

The proof is not a conventional paper. It ships as a Lean library, a body of code that the Lean proof assistant checks step by step. Reports put the library at 14 archive parts holding 175,187 Lean 4 files, or about 17 million lines. The note itself is a short sketch pointing to that code. No price, product or access tier is involved, since the files are public.

Advertisement

The certification from the Lean community

The Lean and Mathlib communities, endorsed by the Lean FRO and the Mathlib Initiative, said in a statement on November 21 that they have full confidence in the proof. Reports quote them as certifying that both the statement and the proof are correct.

The check went beyond compiling the code. Two independent tools, called comparator and con-leche, compared the theorem Alpöge proved with Mathlib's own definition of the Riemann Hypothesis and found them identical. That matters because a Lean proof is only meaningful if the formal statement matches the problem people care about. Andrew Sutherland, a mathematician at MIT, is also reported to have compiled the result on his own without problems.

Claude's role and Anthropic's position

Alpöge credits Claude with a large share of the work, and commentary on the release says Claude produced many of the ideas and handled much of the technical implementation while he steered. Reports say he consulted Anthropic chief executive Dario Amodei before posting. The coverage I reviewed does not show an announcement from Anthropic's official accounts, and the company has not published its own account of how the proof was produced.

Consequences for mathematicians and rival labs

If the result stands, it is the first resolution of a problem open since 1859 that came out of a human and AI collaboration, and it would give every lab a benchmark for what long-horizon reasoning models can do. Rivals such as OpenAI and Google DeepMind have not commented in the reports available so far.

Formal verification changes how the claim can be judged. A reviewer can run the code instead of trusting a long argument, which shortens the usual checking period. The 17 million lines are still far too large for any person to read in full, so scrutiny will focus on the definitions and the statement-matching step rather than on individual proofs.

What remains unproven or unannounced

Reports disagree on how settled the result is. Some outlets describe it as certified. Others, published earlier, said no independent expert review had been completed. The certification covers formal correctness. It does not supply a human-readable explanation of why the argument works, and no standard paper has been released.

The scope also needs clarifying. Coverage conflicts on how fully the work settles the Generalized Riemann Hypothesis, and Alpöge himself describes the Extended version as future work. The Riemann Hypothesis is one of the Clay Mathematics Institute's Millennium Prize Problems, and the institute's rules call for publication and a waiting period before any award. No prize decision has been announced.

Open questions include how many Claude sessions and how much compute the project consumed, whether Anthropic will release the working logs, and when a peer-reviewed write-up will appear. None has been answered publicly.

Featured image: Pixapopz on Pixabay

Advertisement

Get the week's tech news by email

One email a week with Trendblog's stories on AI, hardware, apps and crypto.

We'll email you to confirm your address. Unsubscribe any time. Privacy policy