Anthropic's Alpoge claims Claude found Riemann proof
Anthropic mathematician Levent Alpoge claimed a proof of the Riemann hypothesis using Claude alongside a 79-page paper on Navier-Stokes finite-time blowup.
By Dillip Chowdary • Nov 20, 2026 • Source: thecondia.com

At a glance
- Author
- Levent Alpoge
- Organization
- Anthropic
- Methods
- Claude and Lean proof assistant
- Riemann status
- 4-page note, 14 Lean files, unverified
- Navier-Stokes status
- 79-page paper, missing certificate
- Clay rules
- 2-year post-publication review period
In this post
Anthropic mathematician Levent Alpoge announced that he used the Claude artificial intelligence model to discover a proof of the Riemann hypothesis, the 160-year-old open problem concerning prime number distribution. Hours earlier, Alpoge and mathematician Tristan Buckmaster published a separate preprint asserting that smooth initial conditions make unforced three-dimensional Navier-Stokes equations blow up in finite time, according to thecondia.com's report.
This report details the released artifacts across Alpoge's personal site and public code repositories, examining the specific claims alongside their missing technical components. It is written for engineers, researchers, and mathematicians tracking automated theorem proving, formal verification standards, and the procedural requirements for historic mathematical prize validation.
01Theta lifts and quarter-million page drafts
Alpoge posted a four-page summary note on his website alongside a formalisation in the Lean proof assistant split across fourteen compressed files. He stated that his mathematical strategy took inspiration from Stepanov's proof of the Riemann hypothesis over function fields. According to Alpoge, Claude suggested the decisive step consisting of a long chain of theta lifts scaled to the square root of prime sizes.
The mathematician noted that no conventional human-readable manuscript exists, as the informal draft exceeds 250,000 pages and he had not mastered every technical segment before publishing. Scientific American reported the claim cautiously, highlighting that Lean formalisations can contain coding bugs and require verification that the theorem was defined correctly. Alpoge acknowledged that the proof is not yet accessible to standard academic scrutiny.
Alpoge indicated that his method extends to the Generalized Riemann Hypothesis and may eventually reach the Extended Riemann Hypothesis with more time. While his note lists acknowledgments to mathematicians Manjul Bhargava, Peter Sarnak, and Maksym Radziwiłł, those citations reflect professional context rather than external validation. The claim rests on whether every nontrivial zero of the zeta function sits on one critical vertical line.
02Smooth initial data and missing certificates
The Navier-Stokes claim arrived as a 79-page manuscript authored by Alpoge and Buckmaster accompanied by a public GitHub repository. The researchers argue that smooth initial data for unforced three-dimensional fluid equations lead to solutions that blow up in finite time. If verified, the paper would formally disprove statements A and B of the Clay Millennium Prize problem while proving statements C and D.
Unlike an earlier September announcement from OpenAI that relied on an external smooth force, this work addresses the unforced problem. Scott Armstrong, a mathematician at New York University and Sorbonne, discussed the result on social media, drawing approximately 33,000 views within hours. Buckmaster stated on Mastodon that the team heavily relied on language models to assemble an unpolished write-up after past priority disputes.
Note: The posted Lean package was patched on 20 November but was not rebuilt after the patch, and the numerical certificate remains unreleased. The paper provides a cryptographic hash for the missing numerical certificate and states that the data will be released soon. Furthermore, the repository specifies that the rigorous interval computation verifying the certificate is not part of the Lean formalisation.
03Mechanical compilation versus human mathematical understanding
Advertisement
Tech Pulse Daily
Get tomorrow's pulse first
Join engineers who read Tech Pulse before stand-up. Free, weekday mornings.
Lean checks each logical step against an established, fixed set of foundational rules and standard axioms. A compilation with standard axioms confirms that the formalised statement logically follows from specified definitions. However, a successful build does not prove that the code matches the intended informal mathematical claim, nor does it guarantee that human mathematicians can understand why the result holds true.
An October discussion on MathOverflow highlighted that an earlier Lean-verified zeta-zero project from Anthropic contained incorrect statements in its human-readable version. Participants noted that models can formalise and prove a headline theorem even when accompanying natural-language explanations contain invalid reasoning. This divergence shows that machine verification can produce correct formal code while leaving human comprehension and narrative exposition unverified.
In September, twenty-five Fields medallists signed a declaration titled A Severe Misalignment of AI in Mathematics. The signatories, including Bhargava, argued that hurried announcements bypass the necessary isolation of new ideas and citation of prior work. They asserted that solving isolated problem benchmarks acts merely as a proxy for the primary mathematical objective of deep intellectual understanding.
04Corporate competition on Millennium Prize demonstration stages
The Navier-Stokes release is part of an ongoing contest between Anthropic and OpenAI over advanced mathematical capabilities. Alpoge and Buckmaster previously disputed priority with OpenAI in September after OpenAI released a forced blow-up result following their public discussions. Commercial artificial intelligence laboratories have increasingly begun using famous open mathematical problems as public demonstration stages for their foundation models.
Alpoge, a Morgan Prize winner and Princeton doctorate supervised by Bhargava, has now produced three major mathematical announcements within a single year. These commercial pressures mean research claims now appear before complete manuscripts, reproducible builds, and peer evaluations are ready. Announcements arrive rapidly on social media and personal websites before the broader mathematical community has time to absorb the underlying technical concepts.
Buckmaster explained that the team could not risk waiting to release their Navier-Stokes findings given past priority tensions across the field. While these competitive incentives do not make the underlying claims invalid, they result in unpolished drafts and missing supporting data. Consequently, external researchers must spend significant effort untangling raw code repositories rather than reviewing fully structured academic publications.
05Mechanical verification steps and Clay Institute review
Validating the Riemann hypothesis claim requires a strict series of mechanical checks by independent mathematicians. Outside researchers must download the fourteen compressed files, build the Lean repository from scratch, and confirm that the final theorem defines the Riemann hypothesis or Generalized Riemann Hypothesis. The community must also audit the codebase to ensure that the proof relies solely on standard foundational axioms.
For the Navier-Stokes claim, researchers must complete several sequential verification tasks:
- Certificate audit Outside mathematicians must inspect the released numerical certificate and verify its cryptographic hash against the paper.
- Interval arithmetic Independent researchers must reproduce the external interval computation that validates the approximate blow-up profile.
- Repository rebuild Contributors must rebuild the patched Lean package and check the code against Charles Fefferman's problem statement.
Official recognition under the Clay Mathematics Institute rules requires a much longer and more rigorous verification process. A proposed solution must be published in a qualifying journal, after which it must undergo a two-year waiting period to achieve general community acceptance. A personal website note, a GitHub repository, and unverified Lean archives do not accelerate that formal prize timeline.
Frequently asked questions
Did Levent Alpoge prove the Riemann hypothesis?
Alpoge published a four-page note claiming a proof with Claude and released fourteen compressed Lean files, but the claim has not yet undergone academic peer review or independent build verification.
What does the Navier-Stokes paper claim?
Alpoge and Tristan Buckmaster published a 79-page preprint arguing that smooth initial data cause unforced three-dimensional Navier-Stokes equations to blow up in finite time.
What technical components are missing from the Navier-Stokes proof?
The proof relies on an unreleased numerical certificate and an interval arithmetic computation that is not included in the Lean formalisation.
What are the Clay Millennium Prize rules for claiming a solution?
A candidate proof must be published in a qualifying journal and achieve general acceptance across the mathematical community for two years before an award can be granted.
Sources
Author
Dillip Chowdary
Writes Tech Bytes coverage of AI, engineering, and the tools that actually ship. Editor of Tech Pulse Daily.
Related on Tech Bytes
Advertisement