Anthropic and Stanford claim competing proofs of the Riemann hypothesis
Two different teams—one using an AI model and one using human researchers—have released drafts solving one of mathematics' most famous open problems.
The short version
- Anthropic's proof is written in the Lean programming language for step-by-step computer verification.
- The Stanford draft was written by Alessandro Masullo and reviewed by Brian Conrad.
- The Riemann hypothesis has been an open problem since 1859 and is linked to prime number distribution.
- A confirmed proof would validate hundreds of existing mathematical theories and impact fields like physics and encryption.
Why it matters
If verified, these proofs would provide perfect control over how prime numbers are distributed, instantly turning hundreds of 'assumed' theories into facts. It also marks a significant milestone in AI's ability to navigate complex, high-level mathematical reasoning.
A dual breakthrough in math
In a dramatic turn for the mathematical community, two independent claims emerged within hours of each other regarding the Riemann hypothesis. Anthropic mathematician Levent Alpöge announced he used the AI model Claude to discover a proof of the Generalized Riemann Hypothesis. This hypothesis, first proposed in 1859, has served as a primary goal for number theorists for a long time.
Just hours after Alpöge's announcement, Stanford University researchers released a competing draft. This second approach was produced entirely by humans, led by mathematician Alessandro Masullo. The sudden timing of the AI announcement forced the Stanford team to release their work earlier than originally planned to keep pace with the news.
Both claims focus on the Generalized Riemann Hypothesis, which extends the original problem to a large family of related equations. Because the problem is so significant, mathematicians are taking both the AI-assisted and human-led efforts seriously. Each team is now facing intense scrutiny to see if their respective methods hold up to rigorous peer review.
How the AI reached the finish line
Alpöge worked closely with Claude over several months, initially using the model to explore a simplified toy version of the problem. He set the direction for the research, but the AI eventually took over the heavy lifting. Alpöge noted that the AI moved so far ahead in complex subfields that he has not yet fully mastered all the surrounding math.
The pair solved the puzzle by extending an idea from mathematician Stepanov, who had addressed a similar question in a different system. This method involves a tool similar to taking derivatives of smooth curves, which usually fails on whole numbers. To bypass this, Alpöge and Claude chained special mathematical transformations together to mimic the effect of a derivative.
The AI provided a key insight by determining that a number sequence of length X requires exactly as many transformations as the square root of X. Because the AI's full argument would span over 250,000 pages, Alpöge released it in the Lean programming language. Lean acts as a digital referee, checking every logical deduction step-by-step to ensure no errors or unproven assumptions exist.
The human-led Stanford draft
The Stanford approach represents a traditional path to solving the problem, relying on human intuition and peer review. Alessandro Masullo wrote the draft, which had been undergoing months of quiet vetting by renowned number theorist Brian Conrad. Conrad provided extensive feedback that is currently being incorporated into the final version of the paper.
Stanford professor Ravi Vakil, who serves as the president of the American Mathematical Society, shared the draft on his blog. He emphasized that this work was completed without any assistance from artificial intelligence. Masullo was originally intended to spend a full year refining the draft at Stanford before making it public.
However, the pressure of the Anthropic announcement led to a hurried release of the current version. Vakil has already noted that several people have flagged potential problems in the proof using various AI models. He stated that all issues must be resolved to the satisfaction of the community before the proof can be accepted as true.
The importance of prime numbers
The Riemann hypothesis is central to understanding prime numbers, which are the indivisible atoms of mathematics. While Euclid proved there are infinitely many primes, their exact positions on the number line appear random. Riemann discovered that the zeros of the Riemann zeta function explain this apparent chaos, acting like instruments in an orchestra to correct estimation errors.
A confirmed proof would mean that the randomness of primes actually follows a strict harmony. This would validate hundreds of mathematical papers that currently assume the hypothesis is true. Many of these theories span across various fields, including high-dimensional geometry and the study of fluid dynamics.
Furthermore, prime numbers are the foundation of modern digital encryption. The spacing of Riemann's zeros has also been linked to the energy levels inside atomic nuclei and random particle motion. Solving this problem provides a roadmap for understanding these fundamental physical and digital systems.
What comes next for the field
The mathematical community is now entering a period of intense verification for both claims. Alpöge plans to write a human-readable version of the AI proof over the coming months to make it accessible to other researchers. Meanwhile, Vakil is urging mathematicians to quickly review Masullo's draft and translate it into systems like Lean to close any remaining gaps.
This situation highlights a shift in how research is conducted, as models like Claude have moved from unreliable tools to systems that break through deep roadblocks. Some mathematicians expressed concern that an AI capable of solving such a fundamental problem could signal a major shift in the role of human researchers. The success of these proofs would reshape the landscape of number theory overnight.
Despite the high stakes, Alpöge remains optimistic about the progress. He describes the breakthrough as a dawn of something greater for science. Whether the final proof comes from a human pen or an AI's code, the result will likely be one of the most significant achievements in modern mathematics.
What to watch
- Whether the Lean code for the AI proof passes full verification.
- If the Stanford team can resolve the issues flagged by other mathematicians.
- How the math community reacts to the AI's ability to solve 'Millennium Problems'.
Try it
Look up the 'Riemann zeta function' to see the basic map mathematicians use to study prime numbers.
Sources and links
- ReportingThe Decoder: the article we sourced
Read the original at The Decoder
First reported by The Decoder. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.
Liked it? Say why
Add a line for other readers.
No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.