---
title: Does a Riemann hypothesis proof break encryption? | Vaultan
description: Two Riemann hypothesis proofs appeared on 20 November 2026; one is now machine-certified. What was claimed, what is checked, and why encryption is safe.
url: https://vaultan.app/topics/riemann-hypothesis-proof-encryption/
---

# Does a proof of the Riemann hypothesis break encryption?
No. A proof that the Riemann hypothesis is true would not give anyone a way to break RSA, elliptic-curve cryptography, AES or password hashing. Two proofs were announced on 20 November 2026. On 21 November the Lean and Mathlib communities certified that the computer-checked proof of one is correct; problems have been reported in the other, and neither has been through peer review. This page sets out what was claimed, what has been checked, and what a proof would and would not change.
Published 20 November 2026, 23:55 UTC. Updated 21 November 2026, 23:45 UTC. Vaultan has no connection to either claim and takes no position on whether either proof is correct.

## What we know as of 21 November 2026, 23:45 UTC

- A [four-page note](https://alpo.ge/remark.pdf) posted on 20 November on the website of Levent Alpöge, whom [TechCrunch](https://techcrunch.com/2026/11/20/ai-claimed-a-solution-to-the-167-year-old-riemann-hypothesis-and-a-human-did-too/) describes as a mathematician employed by Anthropic, says he believes he has, working with Claude, found a proof of the Riemann hypothesis and of the generalized Riemann hypothesis (GRH). What was released is code for the Lean proof checker, not a paper.
- MIT mathematician Andrew Sutherland [reports](https://math.mit.edu/~drew/grh/) that all 175,187 files of that code compile with no errors, that its main theorems use only Lean's three standard axioms, and that several separate proof checkers accept it. A [public log of the checks](https://github.com/nasqret/riemann-rev) describes the proof as machine-checked and not yet refereed.
- On 21 November a [joint statement](https://leanprover-community.github.io/blog/posts/riemann-hypothesis/) by the Lean and Mathlib communities, endorsed by the Lean FRO and the Mathlib Initiative, certified that both the statement and the proof of the Riemann hypothesis are correct. It says the theorem proved matches the definition in Mathlib, that dozens of people rebuilt the code without errors, and that the proof passed Lean's own kernel and several independent checkers. The statement covers the Riemann hypothesis and does not mention GRH. There is still no paper or human-readable explanation, and the statement looks ahead to work that will explain the new ideas in the proof.
- Separately, Stanford mathematician Ravi Vakil [posted a draft](https://math216.wordpress.com/2026/11/20/hurried-announcement-alessandro-masullo-grh/) by Alessandro Masullo that proposes a proof of GRH, written with no AI. He says Brian Conrad has examined it for months, and asks others to check it. In an update added about eight hours later, Vakil writes that several people have reported problems with the proposed proof, found with the help of AI models, that every issue has to be resolved to the community's satisfaction before it is accepted, and that he makes no claim that the strategy is correct.
- Neither proof has been peer reviewed or published in a journal, and the Clay Mathematics Institute still [lists the Riemann hypothesis as unsolved](https://www.claymath.org/millennium/riemann-hypothesis/).
Update log (newest first)

- 21 November 2026, 23:45 UTC: added the joint statement by the Lean and Mathlib communities certifying the Lean proof of the Riemann hypothesis, and Ravi Vakil's update that problems have been reported in Masullo's draft. Correction: the previous version said Vakil had posted nothing further; his update had already been added when that was written.
- 21 November 2026, 06:10 UTC: added what has been checked since first publication: Andrew Sutherland's report on the Lean build and axioms, a public log of further checks, and statements in the Lean community's public discussion. Removed a second-hand passage about press reports and the summary site it relied on. Cited original papers in place of most Wikipedia links. Corrected two details about the note. Shortened the page.
- 20 November 2026, 23:55 UTC: first published.

## Quick answers

### Has the Riemann hypothesis been proved?
A computer-checked proof exists, and on 21 November the Lean and Mathlib communities certified that it is a correct proof of the Riemann hypothesis as Mathlib states it. What mathematicians usually mean by proved is still to come: there is no paper, no human-readable explanation and no peer review yet. A separate human-written draft proof of GRH has had problems reported in it that are not yet resolved.

### Does a proof of the Riemann hypothesis break RSA or other encryption?
No. Knowing that the hypothesis is true gives no method for factoring large numbers or solving discrete logarithms, which is what breaking RSA, Diffie-Hellman or elliptic-curve cryptography would take.

### Are AES-256 and password hashing affected?
No. AES, SHA-2, HMAC, PBKDF2 and HKDF are not built on prime numbers or on conjectures about them, so a proof about the distribution of primes does not touch them.

### What would a proof of the generalized Riemann hypothesis change for computing?
It would turn "true if GRH holds" results into theorems, such as Miller's 1976 deterministic primality test. That test is far slower than the ones software uses, so nothing is likely to switch to it.

### What is the real long-term threat to public-key encryption?
A large quantum computer running Shor's algorithm, which would factor numbers and compute discrete logarithms efficiently. NIST approved three post-quantum standards against it, FIPS 203, 204 and 205, in August 2024, and that threat does not depend on the Riemann hypothesis.

### When will we know whether either proof is correct?
For the Lean proof, checking that the code compiles, proves the right statement and passes independent checkers is done, according to the Lean and Mathlib communities. Understanding the argument and agreement among mathematicians take longer, and the Clay prize rules require at least two years after publication in a refereed outlet. For Masullo's draft, the problems reported have to be resolved first.

## What was claimed
The first claim is a short, unsigned [note](https://alpo.ge/remark.pdf) on Levent Alpöge's website. Its author writes that he believes he has found a proof of the Riemann hypothesis while working with Claude, credits Claude with much of the technical work, and says the method extends to GRH and, with time, probably to a broader version. He released a 14-part archive of Lean code and no paper. He writes that his working document, more than 250,000 pages, was unreadable and not fit to post, that the Lean code should settle anything his summary gets wrong, and that he will write the argument up properly.
The archive holds [175,187 Lean files](https://github.com/Hertaspace/alpoge-rh-lean-verification), about 17.4 million lines. Its main theorem uses the definition of the Riemann hypothesis in Mathlib, the standard Lean mathematics library. Four further theorems state the hypothesis for Dirichlet L-functions with [definitions written inside the project](https://github.com/Deicyde/riemann-hypothesis).
The second claim is a draft by Alessandro Masullo, [posted by Ravi Vakil](https://math216.wordpress.com/2026/11/20/hurried-announcement-alessandro-masullo-grh/) on his blog the same day. Vakil writes that Brian Conrad has examined it for months and it looks good so far, and that the plan had been a year of polishing before any announcement. An update to the post about eight hours later says several people have since reported problems with the proposed proof. [TechCrunch](https://techcrunch.com/2026/11/20/ai-claimed-a-solution-to-the-167-year-old-riemann-hypothesis-and-a-human-did-too/) reports that Masullo calls it twelve years of work.
These are two separate claims. One is a machine-checkable artifact that has now passed its formal checks and has no explanation a person can read. The other is a readable draft with no formal check, in which readers have reported problems. Neither has been peer reviewed.

## The Riemann hypothesis and its generalization, briefly
The Riemann zeta function encodes how the prime numbers are spread among the whole numbers. In 1859 Bernhard Riemann proposed that all its "nontrivial" zeros lie on the vertical line where the real part is 1/2, which would pin down how far the primes stray from their average spacing ([Clay Mathematics Institute](https://www.claymath.org/millennium/riemann-hypothesis/)). Computers have checked every zero up to height 3·1012 ([Platt and Trudgian](https://arxiv.org/abs/2004.09765)). The problem carries a US$1 million Millennium Prize.
The generalized Riemann hypothesis makes the same claim for the Dirichlet L-functions, which describe how primes are spread among arithmetic progressions such as 3, 7, 11, 15 and so on. The ordinary hypothesis is the simplest member of this family, so a proof of GRH would also prove it. A broader version for number fields is often called the [extended Riemann hypothesis](https://en.wikipedia.org/wiki/Generalized_Riemann_hypothesis); Alpöge's note presents that one as an expectation.

## What a proof would not change
Breaking RSA means factoring a large number into its two primes; breaking Diffie-Hellman or elliptic-curve cryptography means solving a discrete logarithm. Both are [generally thought to be hard](https://arxiv.org/abs/quant-ph/9508027) on an ordinary computer. GRH describes how primes are distributed and is not a method for either task, and none of its known consequences below is one.
Number theorists have also treated GRH as a working assumption for decades, and results about algorithms have been stated "if GRH holds" since [Miller in 1976](https://www.sciencedirect.com/science/article/pii/S0022000076800438). A proof would confirm what those results assumed, not overturn anything security depends on.
The usual worry is that new ideas inside a proof might one day lead to new algorithms. That is speculation: neither announcement claims a faster way to factor or to compute discrete logarithms, and by [TechCrunch's account](https://techcrunch.com/2026/11/20/ai-claimed-a-solution-to-the-167-year-old-riemann-hypothesis-and-a-human-did-too/) mathematicians have not yet fully understood the proof.

## What a proof of GRH would change
Several results in computational number theory are proved only on the assumption that GRH is true. A proof would make them unconditional:

- Primality testing. In 1976 Gary Miller [showed](https://www.sciencedirect.com/science/article/pii/S0022000076800438) that, if the hypothesis holds, primality can be tested in polynomial time without randomness, and Eric Bach later [showed](https://www.ams.org/journals/mcom/1990-55-191/S0025-5718-1990-1023756-8/) that a composite n always has a witness no larger than 2(ln n)2. The [AKS test](https://annals.math.princeton.edu/2004/160-2/p12) has done the same with no assumption since 2002, and software uses fast randomized tests instead. Our [study of 12 libraries](https://vaultan.app/research/prime-testing-grh/) measures the difference.
- Finding special numbers without luck. Under the hypothesis every proper subgroup of the integers modulo m misses a number of size about (log m)2 ([Bach 1990](https://www.ams.org/journals/mcom/1990-55-191/S0025-5718-1990-1023756-8/)). So a small quadratic non-residue always exists, which lets the Tonelli-Shanks square-root method run in [guaranteed polynomial time](https://en.wikipedia.org/wiki/Generalized_Riemann_hypothesis), and [similar results](https://www.ams.org/journals/mcom/1992-58-197/S0025-5718-1992-1106981-9/) cover primitive roots. In practice software tries random candidates and succeeds almost at once.
- Proofs about running time. The class-group factoring method suggested by Seysen was shown by A. K. Lenstra to run in its expected time if GRH holds, and Lenstra and Pomerance [later removed the assumption](https://www.ams.org/jams/1992-05-03/S0894-0347-1992-1137100-0/). A proof of GRH would settle how such running times are proved. It would not make any method faster.
- Where primes must appear. Under the Riemann hypothesis the gap after a prime p is at most about √p·log p, a result [due to Cramér](https://en.wikipedia.org/wiki/Riemann_hypothesis). Under GRH there are [explicit bounds](https://arxiv.org/abs/1309.3595) for the first prime in an arithmetic progression, and the broader hypothesis gives [similar bounds for number fields](https://www.ams.org/journals/mcom/1990-55-191/S0025-5718-1990-1023756-8/). These bound how long a search for a prime of a required form can take.
In each case "this works if GRH is true" becomes "this works", and no attacker gains a capability they did not already have.

## Encryption that does not involve primes at all
Much of the encryption that protects stored data is symmetric. [AES](https://csrc.nist.gov/pubs/fips/197/final) transforms 128-bit blocks under a 128, 192 or 256-bit key, and SHA-256 and the constructions built on it (HMAC, PBKDF2, HKDF) likewise apply fixed operations to blocks of bits. None rests on a number-theory problem, so a theorem about primes says nothing about them.
Vaultan is one example. Its [published security design](https://vaultan.app/security/design/) encrypts everything in a vault with AES-256-GCM, turns the unlock pattern into a key with PBKDF2-HMAC-SHA256, and derives subkeys with HKDF-SHA256. A proof of GRH changes nothing about how a vault is protected.

## The real long-term threat: quantum computers
The known route to breaking today's public-key cryptography is [Shor's algorithm](https://arxiv.org/abs/quant-ph/9508027) (1995), which would factor integers and compute discrete logarithms in polynomial time on a large enough quantum computer. In August 2024 NIST [approved three standards](https://csrc.nist.gov/news/2024/postquantum-cryptography-fips-approved) designed to resist it: FIPS 203 for key establishment and FIPS 204 and 205 for signatures. It also [says](https://csrc.nist.gov/projects/post-quantum-cryptography/faqs) AES with 128, 192 or 256-bit keys can continue to be used. That migration, not the Riemann hypothesis, is what matters for long-term security.

## What has been checked, and what is still open

- The build and the axioms. [Sutherland reports](https://math.mit.edu/~drew/grh/) a complete build with no errors, and that the five main theorems use only propext, Classical.choice and Quot.sound, the three standard Lean axioms. The [joint statement](https://leanprover-community.github.io/blog/posts/riemann-hypothesis/) of 21 November says dozens of people reported the same clean build. A clean build means Lean's small trusted core accepts a proof of each theorem as stated.
- Independent checkers. The [joint statement](https://leanprover-community.github.io/blog/posts/riemann-hypothesis/) says the exported proof was accepted by Lean's official kernel, by [con-leche](https://github.com/leanprover/con-leche) and con-ron, and by several independent checkers written in other languages, and that the checks ran on an export that cannot run code. [Sutherland](https://math.mit.edu/~drew/grh/), whose page the statement points to for an overview, notes that a pass still depends on those checkers and on Mathlib's definitions being right. His page was last updated on 20 November and lists two of his own checks as still running.
- Whether the statement proved is really the hypothesis. For the Riemann hypothesis this is settled as far as tools can settle it: the main theorem uses Mathlib's own definition, and the [joint statement](https://leanprover-community.github.io/blog/posts/riemann-hypothesis/) says two tools confirmed that the theorem in the library matches it. The four statements about Dirichlet L-functions use short definitions copied from the project, [so readers still have to compare them](https://github.com/Deicyde/riemann-hypothesis) with the standard ones, and the joint statement does not cover them.
- Cleaning up the code. The code is now mirrored in the [Lean community's GitHub organization](https://github.com/leanprover-community/riemann-hypothesis). In the [discussion](https://leanprover.zulipchat.com/) Morrison called it stylistically poor and written for old versions of Lean and Mathlib, and one [analysis](https://github.com/james-shore/rh-proof-dependencies) finds 256,620 of its 1,061,531 declarations unused. Removing dead code and updating versions are among the next steps discussed.
- Whether anyone can explain it. A formal check says that a proof exists, not why the argument works. In the same discussion Lior Silberman wrote that the Lean community has done what it can for now and that digesting the proof is up to number theorists, and Harald Helfgott noted that the announcer says his own understanding of what the tools did is limited. Alpöge says a proper write-up will follow.
- Expert reading of Masullo's draft. This is the traditional route: specialists read it, look for gaps, and the paper goes to a journal. Readers have reported problems, and [Vakil writes](https://math216.wordpress.com/2026/11/20/hurried-announcement-alessandro-masullo-grh/) that all of them have to be resolved before the proof is accepted.
- The Clay prize. The [rules](https://www.claymath.org/millennium-problems/rules/) require publication in a refereed outlet of worldwide repute, a wait of at least two years after that, and general acceptance by mathematicians. The institute takes no direct submissions, and the [full rules](https://www.claymath.org/wp-content/uploads/2022/03/millennium_prize_rules_0.pdf), adopted in 2018, say nothing specific about formal or AI-assisted proofs.
Related research: [How encryption software tests for prime numbers, and what a proof of GRH would change](https://vaultan.app/research/prime-testing-grh/), with the source code of 12 libraries read and Miller's deterministic test timed.

## Sources

- Note on Levent Alpöge's website, 20 November 2026: [alpo.ge/remark.pdf](https://alpo.ge/remark.pdf). Read 21 November 2026.
- Andrew Sutherland (MIT), "RH Lean proof: verification files", last updated 20 November 2026, 22:05 UTC: [math.mit.edu/~drew/grh](https://math.mit.edu/~drew/grh/). Read 21 November 2026.
- Lean community Zulip, channel "general", topic "Discussion: Alpöge Proves the Riemann Hypothesis", 20 November 2026: [leanprover.zulipchat.com](https://leanprover.zulipchat.com/). Read 21 November 2026.
- Public log of the checks of the Lean proof: [github.com/nasqret/riemann-rev](https://github.com/nasqret/riemann-rev). Read 21 November 2026.
- Lean community mirror of the Lean project: [github.com/leanprover-community/riemann-hypothesis](https://github.com/leanprover-community/riemann-hypothesis); the [con-leche](https://github.com/leanprover/con-leche) checker; dependency statistics by James Shore: [github.com/james-shore/rh-proof-dependencies](https://github.com/james-shore/rh-proof-dependencies); Lean FRO [team page](https://lean-lang.org/fro/team/). Read 21 November 2026.
- Ravi Vakil, "Hurried Announcement: Alessandro Masullo GRH", 20 November 2026, with an update added about eight hours later: [math216.wordpress.com](https://math216.wordpress.com/2026/11/20/hurried-announcement-alessandro-masullo-grh/). Read 21 November 2026.
- Tim Fernholz, TechCrunch, 20 November 2026: [AI claimed a solution to the 167-year-old Riemann hypothesis, and a human did, too](https://techcrunch.com/2026/11/20/ai-claimed-a-solution-to-the-167-year-old-riemann-hypothesis-and-a-human-did-too/). Read 21 November 2026.
- Independent build and axiom check of the Lean project (in progress when read): [github.com/Hertaspace/alpoge-rh-lean-verification](https://github.com/Hertaspace/alpoge-rh-lean-verification). Read 21 November 2026.
- Mirror of the Lean project with a sandboxed Comparator check (in progress when read): [github.com/Deicyde/riemann-hypothesis](https://github.com/Deicyde/riemann-hypothesis). Read 21 November 2026.
- Clay Mathematics Institute, Riemann Hypothesis: [claymath.org](https://www.claymath.org/millennium/riemann-hypothesis/); Millennium Prize [rules](https://www.claymath.org/millennium-problems/rules/) and [full rules (2018)](https://www.claymath.org/wp-content/uploads/2022/03/millennium_prize_rules_0.pdf).
- Dave Platt, Tim Trudgian, "The Riemann hypothesis is true up to 3·1012", Bulletin of the London Mathematical Society: [arXiv:2004.09765](https://arxiv.org/abs/2004.09765).
- Gary L. Miller, "Riemann's hypothesis and tests for primality", Journal of Computer and System Sciences 13 (1976), 300-317: [sciencedirect.com](https://www.sciencedirect.com/science/article/pii/S0022000076800438).
- Eric Bach, "Explicit bounds for primality testing and related problems", Mathematics of Computation 55 (1990), 355-380: [ams.org](https://www.ams.org/journals/mcom/1990-55-191/S0025-5718-1990-1023756-8/).
- Manindra Agrawal, Neeraj Kayal, Nitin Saxena, "PRIMES is in P", Annals of Mathematics 160 (2004), 781-793: [annals.math.princeton.edu](https://annals.math.princeton.edu/2004/160-2/p12).
- Victor Shoup, "Searching for primitive roots in finite fields", Mathematics of Computation 58 (1992), 369-380: [ams.org](https://www.ams.org/journals/mcom/1992-58-197/S0025-5718-1992-1106981-9/).
- Youness Lamzouri, Xiannan Li, Kannan Soundararajan, "Conditional bounds for the least quadratic non-residue and related problems": [arXiv:1309.3595](https://arxiv.org/abs/1309.3595).
- H. W. Lenstra, Jr., Carl Pomerance, "A rigorous time bound for factoring integers", Journal of the American Mathematical Society 5 (1992), 483-516: [ams.org](https://www.ams.org/jams/1992-05-03/S0894-0347-1992-1137100-0/).
- Wikipedia, [Riemann hypothesis](https://en.wikipedia.org/wiki/Riemann_hypothesis) and [Generalized Riemann hypothesis](https://en.wikipedia.org/wiki/Generalized_Riemann_hypothesis), for Cramér's bound on prime gaps, the Tonelli-Shanks remark and the name of the extended hypothesis. The original papers are cited there.
- NIST, FIPS 197, Advanced Encryption Standard: [csrc.nist.gov](https://csrc.nist.gov/pubs/fips/197/final).
- Peter Shor, "Polynomial-Time Algorithms for Prime Factorization and Discrete Logarithms on a Quantum Computer": [arXiv:quant-ph/9508027](https://arxiv.org/abs/quant-ph/9508027).
- NIST, post-quantum standards approved, 13 August 2024: [csrc.nist.gov](https://csrc.nist.gov/news/2024/postquantum-cryptography-fips-approved); and the [post-quantum cryptography FAQ](https://csrc.nist.gov/projects/post-quantum-cryptography/faqs).
- Vaultan security design: [vaultan.app/security/design/](https://vaultan.app/security/design/).Found a mistake? Write to support@vaultan.app and we will correct it and note the change in the update log.

[All topics](https://vaultan.app/topics/)
