Riemann Hypothesis Proof Claimed by Anthropic’s Claude

Dynamic aerial view of ocean waves creating bold blue and white abstract patterns.
Photo by Nothing Ahead on Pexels

Key takeaways

  • Anthropic mathematician Levent Alpöge has claimed a Riemann Hypothesis proof developed in collaboration with the AI model Claude.
  • The evidence is presented as a massive Lean 4 project containing approximately 175,000 files rather than a traditional paper.
  • The claim includes the Generalized Riemann Hypothesis (GRH), which has massive implications for number theory.
  • Experts note that while the Lean code allows for machine-verifiable checking, the proof remains unverified by the mathematical community.

A monumental Riemann Hypothesis proof claim was made on Friday, November 20, 2026, by Anthropic mathematician Levent Alpöge, who asserts he discovered the solution while working alongside Anthropic’s AI model, Claude. Instead of a conventional manuscript, Alpöge released a four-page note and a sprawling collection of machine-checkable code written in the Lean 4 proof assistant. If the claim is validated, it would resolve the most significant open problem in mathematics, a challenge that has remained unsolved since Bernhard Riemann proposed it in 1859.

What happened

On the morning of November 20, 2026, a brief, unsigned PDF titled “remark.pdf” appeared on the personal website of Levent Alpöge. In the document, Alpöge stated his belief that he had found a proof for the Riemann Hypothesis (RH) and its broader counterpart, the Generalized Riemann Hypothesis (GRH). He further indicated that with more time, he expects to extend this success to the Extended Riemann Hypothesis (ERH).

Alpöge explained his unconventional decision to forego a standard research paper. He noted that an unedited, AI-generated written manuscript would be an unreadable “slop,” as his own initial draft had ballooned to over 250,000 pages. To ensure mathematical rigor, he opted to release the proof as a formal Lean 4 project, distributed across 14 compressed archive files. This approach allows computers to verify every logical step line by line, bypassing the potential for human error in reading massive volumes of text.

A vibrant workspace featuring colorful code on computer monitors, ideal for developers.
Photo by Jakub Zerdzicki on Pexels

The announcement triggered immediate discussion within the global mathematics community. Harry Stoltz of NYU’s Courant Institute was among the first to flag the news on social media, where the topic garnered hundreds of thousands of views within hours. While the news spread rapidly, as of Friday evening, no independent expert had confirmed the validity of the proof, and Anthropic had not issued an official corporate statement regarding the discovery.

Who is Levent Alpöge?

Levent Alpöge is a highly regarded number theorist who currently works at Anthropic. He earned his PhD from Princeton University in 2020 under the supervision of Manjul Bhargava. His academic pedigree includes winning the Morgan Prize for undergraduate research at Harvard and serving as a Junior Fellow of the Harvard Society of Fellows. Alpöge has a history of working at the intersection of mathematics and advanced computation, including a 2025 project on Hilbert’s tenth problem and a July 2026 claim regarding the Jacobian conjecture, which was also assisted by Claude.

Why it matters

The Riemann Hypothesis is not merely a theoretical puzzle; it is the cornerstone of modern number theory. Proposed by Bernhard Riemann in 1859, the hypothesis concerns the distribution of prime numbers. While primes appear to occur randomly, Riemann discovered that their distribution is intimately tied to the zeros of the Riemann zeta function, ζ(s).

Specifically, the hypothesis posits that all non-trivial zeros of this function lie on a single vertical line in the complex plane, known as the “critical line,” where the real part is exactly ½. If this is true, it confirms that prime numbers are distributed as regularly as possible, providing the most precise possible error term for the Prime Number Theorem. Thousands of existing mathematical theorems begin with the phrase, “Assuming the Riemann Hypothesis is true…” A formal proof would instantly transform these “conditional” results into unconditional mathematical certainties.

The Riemann Hypothesis in plain English

To understand the stakes, one must understand the relationship between primes and the zeta function. Prime numbers—the building blocks of all integers—do not follow a simple linear pattern. However, they do follow a predictable density. As numbers get larger, primes become scarcer at a rate roughly described by the natural logarithm.

Riemann’s breakthrough was showing that the “fluctuations” or errors in this density are controlled by the zeros of the zeta function. Imagine a musical note; the fundamental frequency is the trend, and the harmonics are the zeros. If the zeros all lie on the critical line, the “music” of the primes is perfectly balanced. If even one zero wanders off that line, it would imply a fundamental irregularity in how primes are scattered across the number line.

A woman analyzes data on a computer screen in a modern office
Photo by ThisIsEngineering on Pexels

Why the Generalized Riemann Hypothesis matters even more

While the standard Riemann Hypothesis focuses on the zeta function, the Generalized Riemann Hypothesis (GRH) extends this logic to a much wider class of functions known as Dirichlet L-functions. These functions are essential for understanding how primes are distributed within arithmetic progressions—for example, how many primes end in the digit 1 versus the digit 7.

If Alpöge’s claim regarding GRH holds, it would settle many questions in computational number theory simultaneously. This includes:

  • Primality Testing: It would provide a proven, deterministic polynomial-time algorithm for testing whether a number is prime, specifically validating Gary Miller’s 1976 test.
  • Quadratic Non-residues: It would guarantee that the smallest non-square modulo a prime is extremely small, which is vital for various point-finding algorithms.
  • Arithmetic Progressions: It would provide near-optimal error terms for counting primes in specific sequences, such as $qn + a$.

By proving GRH, Alpöge is targeting a much larger mathematical structure than just the original Riemann Hypothesis, which is why the mathematical community is treating the claim with such intense scrutiny.

How Claude helped: inside Alpöge’s note

One of the most striking aspects of this story is the explicit role of Anthropic’s Claude AI. Alpöge describes the process not as a tool being used, but as a high-level collaboration. In his note, he writes, “Claude flew, and danced, and sparkled near that finish line.”

According to Alpöge, the mathematical strategy was a hybrid of human intuition and AI-driven technical execution. He began with a simplified version of the problem using “Stepanov’s method,” which had previously been used to prove the hypothesis for curves over finite fields. However, applying this to whole numbers is notoriously difficult because it requires a concept called “multiplicity” that does not translate easily to integers.

Alpöge credits Claude with solving this specific hurdle. The AI reportedly suggested imitating multiplicity through a long chain of “theta lifts” between automorphic forms. This technical leap—proposing a chain of lifts with a length scaling with the square root of the prime size—is what Alpöge suggests was the decisive breakthrough. He admits that while he steered the direction, Claude handled much of the heavy lifting in the realm of higher-rank automorphic forms, an area where Alpöge notes he has limited mastery.

A 41-day sprint: the timeline

Alpöge’s note includes a rough 20-step timeline, compiled by Claude, that tracks the rapid progression from the first idea to the completed GRH claim. The timeline spans just 41 days, from October 6, 2026, to November 16, 2026.

Date Milestone/Step
October 6, 2026 Initial steps involving Stepanov determinants and Iwasawa-tower analogies
October 14, 2026 Work on GGP-period determinants and relative trace formulas
October 19, 2026 Transition to weighted prime sums
November 15, 2026 Intensive period: iterated theta lifts and chain length calculations
November 16, 2026 Completion of the Generalized Riemann Hypothesis (GRH) step
November 20, 2026 Public release of the remark and Lean 4 files

The density of the work is most apparent on November 15, a Sunday, when eight distinct steps were recorded. Alpöge noted that the results finally coalesced late that night, leading him to contact Anthropic leadership to discuss the announcement.

How the Lean proof will be checked

Because the proof is written in Lean 4, the verification process differs significantly from traditional peer review. In a standard review, mathematicians read a paper and attempt to find logical gaps. In a Lean-based review, the primary goal is to ensure the code compiles and that the machine-checked logic is sound.

Verification involves several critical stages:

  1. Compilation: Independent researchers must be able to build the entire project using the specific version of the Lean toolchain (v4.33.1) and the Mathlib library specified in the files.
  2. Axiom Audit: A key concern is whether the proof uses any “unproven” assumptions or hidden axioms. Experts will run commands like #print axioms to ensure the proof relies only on the standard, accepted axioms of Lean.
  3. Statement Matching: Using tools like the leanprover/comparator, researchers will check if the theorem stated in the code (e.g., riemannHypothesis) actually matches the mathematical definition of the Riemann Hypothesis used by the broader community.
  4. The “Sorry” Check: In Lean, the keyword sorry is used as a placeholder for unproven steps. A valid proof must contain zero instances of this keyword.

While the project is massive—containing roughly 175,000 files and 17 million lines of code—the fact that it is machine-verifiable means that once the code is successfully built, the logical correctness of the individual steps is mathematically guaranteed by the Lean kernel.

How mathematicians are reacting

The reaction from the mathematical community has been a mixture of profound excitement and necessary skepticism. While some social media posts have prematurely declared the problem “solved,” most professionals are waiting for the results of the Lean builds.

Thomas Bloom of the University of Manchester expressed surprise at both the AI’s involvement and the sheer scale of the unreadable draft. Elliot Glazer, noting the technical nature of the announcement, suggested that the chance of a fundamental bug in a Lean-level proof is extremely low, provided the initial statement is correct.

However, history provides a reason for caution. Several respected mathematicians, such as Louis de Branges and Sir Michael Atiyah, have previously claimed to solve the Riemann Hypothesis, only to have their proofs fail under scrutiny. The difference this time is the presence of the Lean formalization, which moves the verification from a subjective human judgment to a reproducible computational process.

Does it break encryption or Bitcoin? Cybersecurity and crypto implications

Whenever a breakthrough involving prime numbers is announced, fears immediately arise regarding the security of the internet. The concern is that if the distribution of primes is suddenly understood, the encryption protecting bank accounts, private messages, and cryptocurrencies might be broken.

The short answer is no. A proof of the Riemann Hypothesis or the Generalized Riemann Hypothesis does not provide a way to factor large numbers or solve discrete logarithm problems, which are the foundations of modern security.

Why your data remains secure

  • RSA Encryption: RSA’s security is based on the difficulty of factoring the product of two large primes. The Riemann Hypothesis describes how primes are distributed on average, but it does not provide a shortcut or an algorithm for factoring.
  • Bitcoin and Ethereum: These networks rely on elliptic-curve cryptography (ECDSA and Schnorr). The Riemann Hypothesis has no known bearing on the difficulty of the elliptic-curve discrete logarithm problem.
  • Hashing (SHA-256): Bitcoin mining and address generation rely on hash functions, which are mathematically unrelated to the zeros of the zeta function.

In fact, many cryptographers have spent decades building systems based on the assumption that the Riemann Hypothesis is true. A proof would confirm the stability of those assumptions rather than undermining them.

The real security shift: AI-driven research

The more significant security implication is not the theorem itself, but the methodology. If Claude can assist in solving a 167-year-old mathematical mystery, it could eventually be used to search for new cryptanalytic algorithms. We have already seen glimpses of this; in October 2026, Anthropic researchers claimed an AI-assisted discovery of a new algorithm for the 3-SAT problem. While these developments do not break today’s encryption, they suggest that the “hardness” of mathematical problems used in security must be continuously re-evaluated in an era of AI-driven mathematical discovery.

Aerial view of a modern conference room with tables and people working.
Photo by Mike van Schoonderwalt on Pexels

What it means for you

If you are a student or researcher, this event signals a fundamental shift in how mathematics is conducted. The era of the “lone genius” working in isolation is being augmented—or perhaps replaced—by a model of human-AI collaboration. The ability to manage and verify 175,000 files of formal code will likely become a required skill in the next generation of mathematicians.

If you are an investor in the technology sector, the timing of this announcement—occurring just one day after Anthropic’s Nasdaq listing—is noteworthy. While Alpöge has dismissed any cynical interpretation of the timing, the event underscores the massive market interest in companies that are successfully integrating AI into high-level cognitive tasks.

For the general public, the most important takeaway is that your digital life is not under immediate threat from this news. The “math magic” that protects your credit card and your Bitcoin remains intact. The news is a triumph of human and machine intelligence, not a weapon against digital privacy.

Counterpoints and open questions

Despite the excitement, several significant questions remain unanswered. First and foremost is the “human-readable” problem. Without a traditional paper that a mathematician can read and understand intuitively, the community is forced to rely entirely on the Lean kernel. While the kernel is logically sound, it does not provide the reason why the proof works. A proof that is correct but incomprehensible to humans is a difficult thing for science to absorb.

Second, there is the issue of the “statement mismatch.” As noted by some experts, a common error in formal proofs is proving a statement that is subtly different from the one originally claimed. Until the Lean code is compared against a trusted reference using tools like the leanprover/comparator, the claim of a “Riemann Hypothesis proof” remains an unverified assertion.

Finally, there is the question of credit. If Claude provided the most difficult technical steps, how will the history of mathematics record this achievement? Will Alpöge be the sole winner, or does the discovery belong to the partnership between man and machine?

What happens next

The coming weeks will be a period of intense technical auditing. The mathematical community will focus on three primary signals:

  1. The Build Signal: Can independent researchers successfully compile the 14 archive parts without errors?
  2. The Axiom Signal: Does the code pass a clean #print axioms check, showing no reliance on unproven assumptions?
  3. The Human Signal: Can Alpöge produce a simplified, readable manuscript that explains the “theta lift” logic to other mathematicians?

If these hurdles are cleared, the process moves into the realm of formal peer review and journal publication. If the proof is eventually accepted, the Clay Mathematics Institute’s $1 million Millennium Prize will likely not be awarded until at least 2028, following their required two-year period of scrutiny and general acceptance.

As the mathematical community begins the arduous task of auditing 175,000 files of code, the world waits to see if November 20, 2026, will be remembered as the day the most famous problem in math was finally conquered by a partnership of human and artificial intelligence.

Frequently asked questions

Has the Riemann Hypothesis been proven?

Not officially. While Levent Alpöge has claimed a proof using Claude and Lean 4, the mathematical community has not yet independently verified the results. A claim is only considered a proof once it has survived rigorous peer review and achieved broad acceptance among experts.

What is the Generalized Riemann Hypothesis?

The Generalized Riemann Hypothesis (GRH) is an extension of the original Riemann Hypothesis. While the original focuses on the zeta function, the GRH applies to a broader set of functions called Dirichlet L-functions. Proving GRH would have wide-reaching implications for how we understand the distribution of prime numbers in different patterns.

Did Claude prove the Riemann Hypothesis by itself?

No, the discovery was a collaboration. Alpöge describes himself as the director of the project, setting the strategy and guiding the work, while Claude provided critical technical solutions, particularly regarding the use of theta lifts to solve complex problems in automorphic forms.

Does a Riemann Hypothesis proof break encryption or Bitcoin?

No. The security of RSA, Bitcoin, and other encryption methods relies on problems like integer factorization and discrete logarithms, which are not solved by a proof of the Riemann Hypothesis. The proof confirms how primes are distributed but does not provide a way to crack the codes that protect digital data.

References

Featured image: Photo by Nothing Ahead on Pexels

Leave a Reply