Skip to content
sobolewski.proAI today, every eveningSubscribe

Once a day. Unsubscribe with one click. Privacy policy

Alessandro Masullo: a GRH proof with no AI, on the day of Claude’s

Alessandro Masullo spent nearly 12 years on a GRH proof. His 940-page draft, written without AI, came out the same day as the proof found with Claude. What happened?

4 min read

This article in other languages:Polski

In this article
  1. What did Alessandro Masullo announce?
  2. Why did the draft come out now?
  3. Is the proof complete?
  4. How do the two proofs differ?
  5. FAQ

On November 20, 2026, Ravi Vakil posted a hurried announcement on the math216 blog: Alessandro Masullo has a proposed proof of the Generalized Riemann Hypothesis (GRH), written entirely by a human, with no AI. The post comes with a 940-page draft of the paper. A few hours earlier, Levent Alpöge of Anthropic had announced a proof of the same hypothesis found together with Claude.

What did Alessandro Masullo announce?

The paper is called “The Music of the Primes”. In the acknowledgements Masullo writes that he started it near the end of his first year as a graduate student at Stanford under Brian Conrad, and that it “took nearly 12 years to assemble”. According to the author the method covers all number fields and all unitary Hecke characters, which is far more than the Riemann zeta function alone.

From Vakil’s post we know that Conrad has been looking at the proof for months and has given extensive feedback, which is being incorporated right now, and that it is “so far looking good”. The plan had been different: Masullo was to come to Stanford for a year, polish the text, and announce the result only after enough people had seen it.

In a little over an hour and a half the post collected close to 600 points on Hacker News.

Why did the draft come out now?

Neither Vakil nor Masullo mentions Alpöge by name. Vakil only writes that “this is urgent” and that the best thing now is to put out the current draft. He was writing from a meeting of the American Mathematical Society’s Executive Committee and Board of Trustees, just before an hours-long meeting he was about to chair. Masullo adds in the acknowledgements: “In light of the current climate I am being advised to release the current version as it is.”

The order of events is easy to read, though. According to its metadata the PDF was created on November 20 at 18:13 UTC, a few hours after Alpöge’s note, and in the comments under Terence Tao’s post about AI-generated proofs, Masullo’s paper shows up as “an independent, non-AI proof of GRH”.

The acknowledgements end with a message to the mathematician “who feels discouraged by the pace and tone of the current climate”: “we are still here. We are very much alive. We are not going anywhere.”

Is the proof complete?

The text certainly isn’t, and the author doesn’t hide it. Masullo writes that “the math is complete”, but that filling in some writing remains in several parts of the paper and that the introduction is unfinished. You can see it in the file itself: the items “The main ideas” and “The role of Brian Conrad” are headings with nothing under them yet. He adds that he and Conrad are aware of “several glitches here and there”, which he was fixing as he wrote things up.

Excerpts from a letter to Peter Sarnak, sent in mid-October 2026, stand in for the introduction. Masullo also writes that last Sunday he and Conrad visited Sarnak at the IAS, and that the blackboard discussion lasted four and a half hours.

There is no independent confirmation. Vakil calls the paper “a proposed proof”, says he has “great faith in Brian’s judgement”, and asks others to check it quickly, formalize it, and fill any small gaps that turn up. Tao, after a first look, judges that the manuscript “is closer to Deninger’s strategy towards RH than Connes’” and that “it will take some time to digest”. Reports of alleged errors are already circulating in the Hacker News comments, from commenters who can’t assess them themselves (let’s keep our distance from those).

How do the two proofs differ?

Masullo Alpöge
Form a draft meant to be read, 940 pages Lean code only
AI involvement none (according to Vakil) Claude
Time spent nearly 12 years timeline in the note: October 6 to November 16, 2026
Checking Brian Conrad, for months under way, on the Lean Zulip among other places
Approach (according to Tao) closer to Deninger’s strategy the Langlands program and arithmetic geometry

Each has exactly the weakness the other lacks. Masullo’s paper can be read, but no program has checked it. Alpöge’s proof is checked by Lean, except that Scott Aaronson puts it at 17.4 million lines of code and predicts mathematicians will consider the matter closed only once they have reconstructed the ideas themselves, which will probably take years. Aaronson is assuming here that the proof holds up.

If both papers survive verification, we get a rare case: the same hypothesis settled on the same day by two different routes, once by a human and once with a model. On that same day Alpöge and Tristan Buckmaster also announced a proof of a singularity for the Navier-Stokes equations, likewise written with Claude.

FAQ

Who is Alessandro Masullo?

A mathematician who, according to the acknowledgements in his paper, was a graduate student of Brian Conrad’s at Stanford. Conrad and Ravi Vakil were arranging for him to return to Stanford for a year to finish the text.

Has GRH been proved?

We don’t know yet. As of November 20, 2026 there are two announced proofs, Masullo’s and Alpöge’s, and no independent confirmation of either.

Did Masullo use AI?

According to Vakil, no: “It is all human, with no AI.”

  • AI
  • science
  • Claude

AI today, every evening

The day's biggest AI news and new posts from the blog, every evening. Short, with sources.

Once a day. Unsubscribe with one click. Privacy policy