Complex mathematical and physics equations and diagrams on a blackboard, including graphs, integrals, and geometric shapes.

Anthropic Mathematician Claims Lean-Checked Riemann Hypothesis Proof With Claude’s Help

Anthropic mathematician Levent Alpöge claims a Lean-checked Riemann Hypothesis proof with Claude’s help. Verification is still pending.

In short

Anthropic mathematician Levent Alpöge says he has posted a Lean-formalized proof claim for the Riemann Hypothesis and the Generalized Riemann Hypothesis with Claude’s help. The result is still unverified, making the real story one of scrutiny, not celebration.

  • Levent Alpöge publicly posted a Lean-formalized claim of a proof of the Riemann Hypothesis on Nov. 20, 2026.
  • The claim also says the same argument proves the Generalized Riemann Hypothesis.
  • Independent mathematicians have not yet verified the proof, and Lean experts are still checking it.
  • TechCrunch said Alpöge messaged Dario Amodei the night before the public announcement.
  • The Riemann Hypothesis has been open since 1859 and carries a $1 million Clay Mathematics Institute prize.

Anthropic mathematician Levent Alpöge says he has posted a Lean-formalized claim of a proof of the Riemann Hypothesis, and that the same work also proves the Generalized Riemann Hypothesis. The claim, published on his personal website on Nov. 20, 2026, immediately became a major test case for whether frontier AI can help produce machine-checkable advances on one of mathematics’ most famous open problems.

The announcement matters because it sits at the intersection of three things the research world is watching closely: AI-assisted theorem proving, formal verification in Lean, and the long-running chase for a solution to a Millennium Prize Problem. But the story is not a solved one. According to the reporting, the proof is still unverified by independent mathematicians, Lean experts are still checking compilation and correctness, and Anthropic itself has not endorsed the announcement.

What Alpöge says he proved

Alpöge’s claim is that a proof written with Claude’s help establishes both the Riemann Hypothesis and the Generalized Riemann Hypothesis. OfficeChai reported that the result is linked to a 41-day working timeline, while Beinsure said Alpöge posted the claim publicly and described Claude as part of the process that helped him find the argument.

The distinction between a proof claim and a widely accepted proof is critical here. In mathematical research, a theorem does not usually enter the canon because its author says so; it becomes accepted only after peers can independently inspect the logic. In this case, that scrutiny is just beginning.

“I believe I have, working with Claude, discovered a proof of the Riemann Hypothesis”

— Levent Alpöge, Anthropic mathematician

That sentence captures the posture of the announcement: confident, but not yet validated. The claim extends beyond the original Riemann Hypothesis to the broader Generalized Riemann Hypothesis, making it even more consequential if it survives review.

Why this claim matters

This is not just another AI headline. If Alpöge’s proof holds up, it would be one of the clearest demonstrations yet that a frontier model can contribute to formal mathematics at the highest level, not merely by suggesting heuristics, but by helping generate a proof that can be checked in Lean.

The Riemann Hypothesis has stood since 1859 as one of the most important unsolved questions in mathematics, and TechCrunch noted that it has remained open for 167 years. The Clay Mathematics Institute has tied it to a $1 million prize, which makes any credible breakthrough instantly newsworthy. A verified proof would also reverberate well beyond pure math, because the hypothesis is deeply connected to the distribution of prime numbers.

There is a second reason the claim is drawing attention: it may reveal how AI changes the workflow of mathematical discovery even before it changes the final result. If a model can help an expert navigate a difficult landscape of lemmas, bounds and formal steps, the value may be less about replacing mathematicians than about extending what they can systematically check.

Why it matters for AI

According to the reporting, the result would be a major proof-of-concept for AI-assisted mathematics in 2026, a year in which labs have increasingly used models to search for strategies and formalize proofs. It would also sharpen the question of what counts as “AI progress” in math: generating plausible arguments, or producing code that compiles and proves the intended theorem.

That distinction matters because formal theorem proving is unforgiving. A model can sound persuasive and still be wrong. Lean, the proof assistant involved here, is designed to make those errors visible. The fact that the work is described as Lean-compilable is therefore significant, but only if the compilation and the mathematical meaning both hold up under independent inspection.

What Lean changes

Lean is a proof assistant used to formalize and machine-check mathematical arguments. In practice, that means a theorem must be written in a way the system can verify step by step, rather than in the free-form prose mathematicians usually publish. Beinsure reported that Lean experts are still checking the code compilation and correctness, which is exactly the kind of review that will determine whether this story remains a claim or becomes a result.

TechCrunch added an important nuance: the result is described as Lean-compilable, not yet human-intelligible. That matters because a proof that a machine can check may still be too opaque for humans to understand comfortably without substantial translation. In other words, the gap between “formally verified” and “conceptually understood” remains real.

How formal proof checking works here

At a high level, the workflow appears to have been: a mathematician develops an argument with help from Claude, encodes it in Lean, and then checks whether the proof assistant accepts it. The human role does not disappear; it changes. The model may assist with search, restructuring, or filling in technical steps, while the mathematician bears responsibility for the mathematical intent.

That is why the verification stage is so important. A Lean file can compile while still failing to capture the theorem the author intended, or it can expose a subtle gap that prose might gloss over. This is especially true for a claim as deep as the Riemann Hypothesis, where every hidden assumption matters.

d>

Milestone Date What the reporting says happened
Riemann Hypothesis formulated 1859 The problem was first posed and later became a Millennium Prize Problem.
Timeline begins Oct. 6, 2026 OfficeChai says the first dated step references a Stepanov-style approach and key bounds.
Turning point Nov. 15, 2026 The timeline records a long chain of iterated theta lifts and related estimates.
Timeline ends Nov. 16, 2026 The final entry reaches the Generalized Riemann Hypothesis.
Public announcement Nov. 20, 2026 Alpöge publicly posts the claim and formalized proof.

How the proof narrative unfolded

OfficeChai described the work as spanning 41 days, with the earliest timeline entries dated Oct. 6, 2026 and the final entry dated Nov. 16, 2026. The publication on Nov. 20 then turned that private or semi-private working sequence into a public claim.

According to OfficeChai, Claude also compiled a rough 20-step timeline of the argument. That timeline reportedly moved through a series of technical landmarks: an early Stepanov-style approach, bounds that moved past a threshold of “greater than 1 minus a small error,” a vanishing statement on a special curve, determinants, GGP periods, a relative trace formula, an identity connecting the problem to theta functions, and a weighted prime sum that raised open questions before the final push.

What the timeline suggests

The reporting points to a highly structured proof search rather than a single flash of inspiration. The sequence of dated entries suggests iterative refinement, with some steps becoming more sophisticated as the month progressed. The final leap on Nov. 15 appears to have been a decisive consolidation of earlier estimates rather than a stand-alone trick.

That is one reason the claim is drawing interest beyond the mathematics itself. It offers a rare glimpse into how a large language model can contribute to a long, technical proof search over many days, not just help with isolated algebraic manipulations.

What makes this different from earlier AI-math stories?

This claim differs from many earlier AI and math headlines because it is not only about conjecturing an answer; it is about formalization. In previous cases, the community has often seen models propose useful ideas, draft proof sketches, or improve partial bounds. Here, the claim is that the argument has been encoded in Lean and that the formalized version covers both the Riemann Hypothesis and GRH.

OfficeChai also noted a related context from earlier in the year: a Claude-assisted improvement on a Riemann-related bound, and a cited OpenAI claim involving an 88-hour Navier-Stokes blow-up proof attempt. Those examples matter because they show a broader race among labs and researchers to use AI in very hard mathematical and physical problems. But the comparison also highlights the central issue: the field has seen many promising results, and far fewer verified ones.

TechCrunch reported another relevant detail: Alpöge and Tristan Buckmaster also released another proof related to Navier-Stokes. That suggests Alpöge’s mathematical work spans more than one major problem domain, although the Riemann claim is the one attracting the sharpest scrutiny.

Background: why the Riemann Hypothesis matters

The Riemann Hypothesis is one of mathematics’ best-known unsolved problems. At its core, it concerns the zeros of the zeta function and, by extension, the distribution of prime numbers. Because primes are foundational to number theory, any proof would ripple through a large part of modern mathematics.

The Generalized Riemann Hypothesis expands that idea to a broader family of L-functions, widening the scope of the claim substantially. That is why the report that Alpöge’s work covers both RH and GRH is so striking. It is not just a narrow advance; it would be a major consolidation of a much larger territory.

The Clay Mathematics Institute’s $1 million prize remains attached to RH, underscoring the prestige of the problem. But prizes only matter if the community accepts the proof, and that process can take time. Beinsure said formal recognition would require two years of community scrutiny, a reminder that the mathematics world moves by verification, not announcement.

How are mathematicians likely to assess the claim?

They will start by checking whether the Lean proof actually compiles and whether every dependency is sound. That is the first gate. The second is whether the theorem formalized in the proof matches the intended statement exactly. The third is whether the proof can be translated into a human-readable argument that specialists can inspect and discuss.

Those checks will not be quick, and the reporting makes clear that the process is already underway. Beinsure said Lean experts are still examining the code. OfficeChai and TechCrunch both emphasized the importance of verification rather than celebration. Beinsure also noted that Anthropic has not endorsed the announcement, which further underlines that this is still an individual claim, not a company position.

“The precise timing is horrible”

— Levent Alpöge, Anthropic mathematician

That remark, reported by TechCrunch, appears to refer to the awkward overlap between the timing of the proof claim and Anthropic’s broader corporate moment. The company became the first publicly traded frontier AI lab the same week, according to TechCrunch. That backdrop guarantees extra attention, but it does not change the burden of proof.

Why the company context matters

TechCrunch’s note that Anthropic became the first publicly traded frontier AI lab the same week adds a distinct layer to the story. Public markets tend to amplify attention around anything that looks like a breakthrough, especially when it involves one of the best-known names in AI. That does not make the math more likely to be correct, but it does affect how widely the claim will travel.

It also raises a governance question. If a researcher at a frontier lab announces a landmark mathematical result, how much of the surrounding institutional credibility travels with the announcement? In this case, Beinsure’s note that Anthropic has not endorsed the claim is the relevant safeguard. The company’s silence signals that the proof is still a researcher-led claim, not an organization-certified result.

By the numbers

Several figures in the reporting help frame the scale of the claim and the scrutiny it now faces.

Figure What it refers to Why it matters
167 years Time since the Riemann Hypothesis was posed Shows the historical weight of the problem
41 days Length of the timeline in OfficeChai’s account Suggests a rapid but structured proof search
20 steps Claude’s rough timeline of the argument Signals a formal, staged proof process
250,000 pages Length of the unpublished human-readable version, per Beinsure Shows how sprawling the explanatory layer may be
$1 mn Clay Mathematics Institute prize for an accepted RH solution Explains the problem’s prestige and stakes
Two years Community scrutiny period mentioned by Beinsure Highlights how long validation could take

One striking number in the report is the claimed 250,000-page length of an unpublished human-readable version. If accurate, that would underscore the practical problem of communicating a formal proof that is technically valid but too long or intricate for a conventional exposition. It also suggests why the community’s job is only beginning.

What happens next

The immediate next step is independent verification. Mathematicians and Lean specialists will need to confirm that the proof compiles cleanly, that the theorem statement is correct, and that no hidden gap remains between the formal object and the mathematical claim. Until that happens, the result should be treated as a promising announcement rather than an accepted theorem.

Watch for three things in the coming weeks and months:

  • Whether Lean experts confirm the code and its mathematical meaning.
  • Whether independent mathematicians can read, reproduce and endorse the argument.
  • Whether a shorter, human-readable account emerges that explains how the proof works.

There is also a broader question of attribution. If the result survives scrutiny, how should credit be divided between Alpöge and Claude? That issue is not cosmetic; it will shape how future AI-assisted mathematics is described, rewarded and replicated.

What should readers watch for?

Readers should watch for whether the community converges on one answer to a simple question: does this proof actually prove what it claims to prove? Until the answer is yes, the story remains an important example of AI-assisted mathematical ambition, but not yet a mathematical breakthrough.

If the claim stands, the implications will be large. If it falls apart, that too will matter, because it will tell researchers where formalized AI assistance is strong, and where it still needs far more human checking than headline language suggests.

Bottom line

Levent Alpöge’s Nov. 20, 2026 announcement is a serious claim of a Lean-formalized proof of the Riemann Hypothesis and the Generalized Riemann Hypothesis, developed with Claude’s help. The reporting suggests the work may be technically elaborate and machine-checkable, but the proof is not yet independently verified, and that verification will decide whether this becomes a landmark or a cautionary tale.

Frequently asked questions

Did Anthropic prove the Riemann Hypothesis?

No. Levent Alpöge, an Anthropic mathematician, publicly posted a claim of a Lean-formalized proof with Claude’s help, but the proof has not been independently verified and Anthropic has not endorsed the announcement.

What is the Generalized Riemann Hypothesis?

It is an extension of the Riemann Hypothesis to a broader family of L-functions. In this story, Alpöge’s claim is notable because it reportedly covers both RH and GRH, which makes the claim broader than a single-problem result.

Why does Lean matter in this proof claim?

Lean matters because it can machine-check mathematical arguments step by step. A proof that compiles in Lean is more than a written sketch, but it still must be checked to ensure it states and proves the intended theorem correctly.

Has the proof been verified by other mathematicians?

No. According to the reporting, the proof remains unverified by independent mathematicians, and Lean experts are still reviewing whether the code compiles and whether the formal argument is correct.

What happens if the proof is accepted?

If the proof is accepted, it would resolve one of mathematics’ most famous open problems and could qualify for the Clay Mathematics Institute’s $1 million prize, subject to formal recognition rules and community scrutiny.

Share this 🚀