Did Claude Really Solve the Riemann Hypothesis? What We Know
Levent Alpoge says he and Claude proved the Riemann Hypothesis in six weeks. Here's the claim, the evidence, and what's still unverified.

What is the actual claim?
On November 20, mathematician Levent Alpoge posted a short note on his personal website stating that he, working with Anthropic’s Claude, had discovered a proof of the Riemann Hypothesis, a problem that has resisted every attempt at a solution since 1859. He went further, saying the same method extends to the generalized Riemann Hypothesis and, with more time, the extended Riemann Hypothesis too. As of this writing, the proof has not been accepted or confirmed by the broader mathematical community. It has been released as a formal Lean proof, and mathematicians are actively checking it. That verification process is ongoing, not finished.
TL;DR
- Levent Alpoge, a number theorist who left Harvard for Anthropic in 2024, says he and Claude produced a proof of the Riemann Hypothesis in roughly six weeks, with the key breakthrough reportedly landing around November 16.
- The proof was released as Lean code, a formal language that lets a computer mechanically check every logical step, rather than as a traditional journal paper, and reports describe it running into the millions of lines.
- Independent verification is still underway: mathematicians including Andrew Sutherland at MIT have compiled and reviewed parts of the Lean code, and online math communities are actively combing through it, but no formal peer-reviewed confirmation exists yet.
- Alpoge previously used Claude to find a counterexample to the Jacobian conjecture in July, an open problem from 1939, making this his second major AI-assisted result in a few months.
- The announcement landed in the same week as Anthropic’s IPO, which Alpoge himself noted created an awkward timing problem around disclosure rules.
- A second, independently authored Riemann proof attempt (by Alessandro Micheletti, a roughly 940-page paper without a Lean proof) surfaced around the same time, which is common when a long-standing problem suddenly becomes “live.”
- The claim matters beyond bragging rights because huge areas of number theory already assume the Riemann Hypothesis is true; proving it would convert those results from conditional to confirmed.
Remy AI Engineer Certification
Companies want to build their own software.
Few people can build it right.
Become the one they trust.
- Access
- Audit trails
- Security
- Compliance
What is the Riemann Hypothesis, in plain terms?
Prime numbers (2, 3, 5, 7, 11, 13…) can only be divided evenly by 1 and themselves. Every other whole number is “composite,” built by multiplying primes together, which is why primes are often called the atoms of arithmetic. Zoomed in, their spacing looks essentially random.
In 1859, German mathematician Bernhard Riemann proposed that this apparent randomness isn’t random at all. He connected the distribution of primes to a function called the Riemann zeta function, and conjectured that all of the function’s nontrivial zeros line up along a single specific line when plotted on the complex plane. That orderly pattern, if true, would explain exactly how primes are distributed. Nobody has ever proven it, and nobody has disproven it either. It’s one of the seven Millennium Prize Problems, each carrying a one-million-dollar reward from the Clay Mathematics Institute, and until now it had an open bounty with zero solutions.
Why does this problem matter so much?
The honest answer is that mathematicians have been building on top of it for over a century without knowing if the foundation holds. A huge number of theorems in number theory begin with the phrase “assuming the Riemann Hypothesis is true.” Those results aren’t guesses, they’re rigorous proofs, but they’re conditional on something nobody could confirm. If Alpoge’s proof holds up, all of that conditional math becomes unconditional, overnight. It’s less like discovering a new building and more like finally inspecting the foundation that a thousand buildings were already standing on.
Who is Levent Alpoge?
Alpoge is a trained number theorist, previously at Harvard, who according to people who studied alongside him was considered one of the most talented mathematicians in his cohort as early as 2010. He left academia to join Anthropic in 2024. This isn’t his first headline-grabbing AI-assisted result: in July, working with Claude, he found a counterexample disproving the three-dimensional Jacobian conjecture, a problem open since 1939. He’s also been connected to the discussion around the Navier-Stokes existence and smoothness problem (another Millennium Prize problem), an area where OpenAI separately published AI-assisted work using a large-scale, many-agent compute approach.
Alpoge has a habit of publishing results directly on X and his personal site rather than through traditional peer-reviewed journals, which has made his announcements both faster to reach the public and slower to be formally vetted.
How was the proof actually verified, and what’s the status now?
| Vendor | Tier | Risk | ApprovalNEW |
|---|---|---|---|
| Northwind Supply | Tier 1 | Low | Approved |
| Acme Logistics | Tier 2 | Medium | Approve |
| Helios Data | Tier 1 | High | Pending |
| Granite Freight | Tier 3 | Low | Approved |
| Pinecrest Labs | Tier 2 | Low | Approve |
This is the part that separates Alpoge’s claim from a typical “I solved it” post. Instead of (or alongside) a written argument, he released the proof formalized in Lean, a programming language built specifically so that mathematical proofs can be checked step by step by a computer rather than a human reader. If a single logical step is invalid, the Lean code simply won’t compile. Reports describe the formalized proof spanning well into the millions of lines of code, split across multiple downloadable files because of its sheer size, with a full written-out version that would run into hundreds of thousands of pages.
That scale means no single person can read the whole thing in a reasonable time. Instead, verification is happening in public, distributed fashion: mathematicians and the Lean community are independently compiling the code and checking that it holds together logically. Andrew Sutherland at MIT reportedly compiled the Lean code independently and described it as a significant moment. Other mathematicians, including Terence Tao, have been explicit that this result is less “digested” than prior AI-assisted math claims simply because it’s so new. Tao’s caution is worth taking seriously: he’s one of the most respected living mathematicians, and his position is essentially “give it time,” not “this is confirmed” or “this is false.” As of now, nobody credible has published evidence that the proof is wrong. That’s a meaningfully different situation than most extraordinary claims face in their first 48 hours, but it is still short of formal acceptance by the mathematical community.
Is the timing with Anthropic’s IPO suspicious?
It’s worth addressing because it came up directly in Alpoge’s own account. He finished the work on a Sunday night and immediately messaged Anthropic CEO Dario Amodei. The problem was that Anthropic’s IPO and the days surrounding it are subject to strict disclosure and communications rules for public companies. Announcing that an Anthropic researcher, using Anthropic’s own model, had just cracked a Millennium Prize problem during that quiet period could have created real legal complications. The announcement went out once that window had passed. This explains a delay in timing; it doesn’t by itself say anything about whether the proof is correct or incorrect. Those are separate questions, and only the second one matters mathematically.
What does this mean for AI and mathematics going forward?
Set aside whether this specific proof survives scrutiny. The pattern around it is the more durable story. A forecasting community (Metaculus) had the Riemann Hypothesis pegged for resolution around 2035 as recently as this year. Terence Tao, on record, predicted AI could plausibly help crack it before 2050. Even the most AI-optimistic public predictions were beaten by close to a decade, if this holds.
The broader trajectory is also hard to ignore: models that reportedly struggled with reliable multi-digit arithmetic a couple of years ago have since won medals at international math olympiads and are now credited as collaborators on Millennium Prize-level problems. Whether that reflects genuine mathematical reasoning or an extremely sophisticated form of pattern completion is a real debate, but the practical distinction may matter less than it sounds. A system that can reliably predict the next correct line in a proof nobody has ever written has to be doing something more than memorizing prior text, because there is no prior text to draw from.
Frequently Asked Questions
Has the Riemann Hypothesis officially been solved?
Everyone else built a construction worker.
We built the contractor.
One file at a time.
UI, API, database, deploy.
Not yet, officially. Levent Alpoge has published a proof and a supporting Lean formalization, and early independent checks have been encouraging, but formal acceptance by the mathematical community (the kind that follows extended peer review) has not happened. Treat it as a strong, actively-verified claim rather than a settled result.
What is a Lean proof, and why does it matter here?
Lean is a programming language and proof assistant that checks mathematical arguments line by line using formal logic. If a proof compiles in Lean, every step is logically valid according to the system’s rules, which gives much stronger confidence than a plain-text argument that only humans have reviewed. It doesn’t guarantee the proof’s premises or framing are meaningful, but it rules out hidden logical gaps.
Who is Levent Alpoge and why does his track record matter?
Alpoge is a number theorist formerly at Harvard who joined Anthropic in 2024. He produced a separate AI-assisted result in July, a counterexample to the Jacobian conjecture, before this Riemann Hypothesis claim. That prior result gives some context for evaluating his credibility, though it doesn’t substitute for independent verification of this specific proof.
Why do so many math results already assume the Riemann Hypothesis is true?
Because mathematicians found the conjecture so compelling and well-supported by numerical evidence that they built further theorems on top of it, labeling them conditional. If the hypothesis is proven true, those theorems become unconditionally proven. If it were ever disproven, a large amount of number theory would need to be revisited.
Did another mathematician also claim a Riemann Hypothesis proof around the same time?
Yes. A separate proof attempt, authored independently and reportedly around 940 pages without an accompanying Lean formalization, surfaced around the same period. Competing or overlapping claims tend to appear once a long-dormant problem becomes active, since researchers who’d been quietly working on it for years rush to establish priority.

