Skip to main content
Number TheoryRiemann Hypothesis· 6 min read· in Science

Anthropic Mathematician Claims AI-Assisted Proof of the 167-Year-Old Riemann Hypothesis

Anthropic researcher Levent Alpöge has published a claimed proof of the Riemann hypothesis, generated with the AI model Claude and formalized in the Lean programming language. The 167-year-old mathematics problem remains unverified by independent experts, who are now examining the massive code release.

By Ishani Patel

Share this story

Listen

On Friday, November 20, 2026, Anthropic mathematician Levent Alpöge announced that he and the artificial intelligence model Claude have discovered a proof of the Riemann hypothesis. The 167-year-old conjecture is widely considered the most famous unsolved problem in pure mathematics.[1][5]

Alpöge published a four-page note on his personal website detailing the claim, alongside a massive formal verification file. He stated that the AI system also resolved the Generalized Riemann Hypothesis, extending the breakthrough across a broader class of mathematical functions.[2][4]

"I believe I have, working with Claude, discovered a proof of the Riemann Hypothesis," Alpöge wrote in the document. The announcement immediately dominated academic and technology discussion boards, drawing intense scrutiny from number theorists worldwide.[2]

The Riemann hypothesis, first proposed by German mathematician Bernhard Riemann in 1859, concerns the distribution of prime numbers. It posits that the non-trivial zeros of the Riemann zeta function all lie on a specific vertical line in the complex plane, known as the critical line.[1][5]

Proving the hypothesis carries a $1 million reward from the Clay Mathematics Institute as one of the seven Millennium Prize Problems. More importantly, a verified proof would immediately confirm hundreds of dependent theorems in number theory and cryptography that assume the hypothesis is true.[1]

A Machine-Checked Argument

Rather than releasing a traditional academic paper, Alpöge published the proposed proof entirely in Lean. Lean is a specialized programming language and interactive theorem prover that checks every logical step of a mathematical argument by machine.[2][4]

The code was released as a compressed archive split across 14 separate volumes. According to early examinations by computational mathematicians, the unpacked files total approximately 1.3 gigabytes of formal logic, an unprecedented size for a single mathematical proof.[2][4]

Alpöge explained that translating the AI's work into a standard human-readable format was currently impossible. "It being obviously inappropriate to attempt to post some massive unreadable travesty of 'slop' (I of course amused myself, and hit 250,000+ pages.), I decided to only present Lean," he wrote.[2][4]

The reliance on Lean shifts the immediate burden of verification from human peer review to computational compilation. If the Lean kernel successfully compiles the 14-part archive without errors, it guarantees that the logical steps are mathematically sound.[4]

However, experts caution that a compiling Lean project does not automatically mean the Riemann hypothesis has fallen. Independent mathematicians must still verify that the theorem Alpöge formalized in the code faithfully represents the actual Riemann hypothesis, without introducing hidden or flawed assumptions.[2][3]

The Role of Artificial Intelligence

The announcement marks a dramatic escalation in the use of large language models for frontier mathematical research. Alpöge described a collaborative process where he guided the high-level strategy while the AI executed the complex technical maneuvers.[4]

He noted that the model's capabilities accelerated as the proof took shape. According to Alpöge's timeline, he steered the investigation toward a specific mathematical setting, but "Claude far outpaced me near the finish line."[4]

The approach reportedly utilized a prime-weighted von Mangoldt sum and a chain of mathematical constructions known as theta lifts. These techniques allowed the human-AI team to enforce square-root bounds on error terms, a critical requirement for wrangling the zeta function's behavior.[2]

Alpöge acknowledged that the resulting argument is likely inefficient, given the speed of the collaboration. He wrote that he "may have baked lots of strange suboptimal choices into the argument," predicting that human mathematicians will eventually distill the massive Lean file into much shorter, elegant proofs.[2]

Building on August's Breakthrough

The November 20 claim follows a highly publicized milestone achieved by Anthropic just three months earlier. In August 2026, the company revealed that an unreleased research version of Claude had made unexpected progress on the exact same problem.[4]

During that earlier experiment, Anthropic staff member Jarred Sumner prompted the model to attempt the Riemann hypothesis. Over a day and a half, the system coordinated approximately 60 subagents and generated 31 million tokens of reasoning to explore the mathematical space.[4]

While that August run did not prove the full hypothesis, it successfully raised the proven lower bound of zeros on the critical line from 41.6 percent to 67.2 percent. That advance was validated by external zeta-function specialists Brian Conrey and Dan Goldston.[4]

In August 2026, an earlier run by Claude successfully raised the proven lower bound of zeros on the critical line to 67.2 percent.

Moving the bound past two-thirds was a historic achievement, but mathematicians noted at the time that the remaining 32.8 percent could contain infinitely many exceptions. The new claim by Alpöge asserts that Claude has now closed that final, infinitely complex gap.[4]

The Verification Challenge

The mathematics community has reacted with a mix of intense excitement and deep skepticism. The Riemann hypothesis has a long history of high-profile claimed proofs that ultimately collapsed under rigorous peer review.[1][2]

Because there is no traditional manuscript to read, number theorists cannot simply scan the document for obvious conceptual errors. They must either learn to navigate the massive Lean codebase or wait for computational specialists to confirm the architecture of the formalization.[2][4]

Anthropic itself has maintained a quiet stance following the publication. The company did not issue an immediate press release on November 20, leaving Alpöge's personal website as the sole source of the announcement and the code.[4]

Alpöge addressed the sudden nature of the release in his note, acknowledging that the timing might raise eyebrows. He insisted that the schedule was dictated purely by the mathematics, stating that "the obvious cynical view... is false."[4]

If the Lean proof compiles and the axioms are verified, the result will fundamentally alter the trajectory of pure mathematics. It would definitively prove that prime numbers are distributed with an intricate, predictable order, rather than appearing randomly along the number line.[1][5]

Beyond number theory, a verified AI-generated proof of this magnitude would signal a paradigm shift in scientific discovery. It would demonstrate that artificial intelligence systems can now resolve century-old intellectual roadblocks that have defeated generations of human geniuses.[1]

Illustration: The immediate burden of verification now shifts to computational compilation, as experts attempt to run the massive Lean codebase.

The Future of Mathematical Discovery

The integration of AI into this level of theoretical work raises profound questions about the future role of human mathematicians. If a machine can generate 250,000 pages of logical deduction to solve a Millennium Prize problem, the human skillset may shift from proving theorems to formulating the right questions.[4]

Alpöge’s own track record lends significant weight to the claim, preventing it from being dismissed as an amateur error. Earlier in 2026, he successfully utilized Claude models to find a three-dimensional counterexample to the Jacobian conjecture, establishing his expertise in AI-guided mathematics.[4]

As the weekend approaches, computational researchers are actively working to compile the 14-part archive. Until that process finishes and the underlying definitions are audited, the 167-year-old Riemann hypothesis officially remains an open question.[2][3]

The sheer scale of the 1.3-gigabyte proof highlights a growing divide between human comprehension and machine verification. While a computer can validate millions of logical steps in hours, understanding the underlying mathematical narrative may take human researchers years of dedicated study.[4]

The sheer scale of the 1.3-gigabyte proof highlights a growing divide between human comprehension and machine verification.

For now, the massive archive sits on servers around the world, waiting for machines and mathematicians to render a verdict. The resolution of the 1859 conjecture currently depends entirely on whether a computational kernel will return a simple checkmark.[4]

Key points

  • Anthropic mathematician Levent Alpöge claims to have proven the 167-year-old Riemann hypothesis using the AI model Claude.
  • The proposed proof was released entirely as a massive 14-part formal verification codebase in the Lean programming language.
  • The claim remains unverified by independent experts, who must now compile and audit the 1.3-gigabyte archive.
  • The announcement follows an August milestone where Claude successfully raised the proven lower bound of zeros on the critical line to 67.2 percent.

Unanswered questions

  • Whether the 14-part Lean archive successfully compiles without errors.
  • Whether the formal definitions in the code faithfully represent the actual Riemann hypothesis without hidden assumptions.
  • How long it will take for independent mathematicians to translate the machine logic into a human-readable format.

How we got here

  1. 1859

    German mathematician Bernhard Riemann proposes the Riemann hypothesis regarding the distribution of prime numbers.

  2. 2000

    The Clay Mathematics Institute designates the Riemann hypothesis as one of the seven Millennium Prize Problems, carrying a $1 million reward.

  3. August 2026

    Anthropic announces that an unreleased version of Claude raised the proven lower bound of zeros on the critical line to 67.2 percent.

  4. November 20, 2026

    Levent Alpöge publishes a claimed full proof of the hypothesis generated with Claude, releasing the argument as a 14-part Lean codebase.

Computational Mathematicians 50%Traditional Number Theorists 50%
Computational Mathematicians
Focusing on the mechanics of the Lean formalization rather than the mathematical narrative.
Traditional Number Theorists
Maintaining deep skepticism until a human-readable argument is produced.

Perspectives this story doesn't cover

  • Clay Mathematics Institute representatives
  • Independent Lean compilation experts

Sources

Source coverage

5 outlets

2 viewpoints surfaced

Computational Mathematicians 50%Traditional Number Theorists 50%
  1. [1]ClarínTraditional Number Theorists

    Has the Riemann hypothesis been solved? Anthropic mathematician Levent Alpöge claims Claude cracked math's most famous problem

    Read on Clarín →
  2. [2]MLQ.aiComputational Mathematicians

    Anthropic mathematician Levent Alpöge posts a claimed proof of the Riemann hypothesis made with Claude

    Read on MLQ.ai →
  3. [3]Superpower DailyTraditional Number Theorists

    Anthropic mathematician claims Riemann hypothesis proof with Claude, pending expert checks

    Read on Superpower Daily →
  4. [4]CellCogComputational Mathematicians

    Anthropic's Levent Alpöge says he and Claude proved the Riemann Hypothesis and formalized it in Lean. What

    Read on CellCog →
  5. [5]New ScientistTraditional Number Theorists

    Anthropic mathematician claims AI-assisted proof of the Riemann hypothesis

    Read on New Scientist →
Listen

Comments

Stay informed

Every angle. Every day.

Get Science stories with full source coverage and perspective breakdowns, free every day.