Claude Helped an Anthropic Mathematician Claim a Proof of the 167-Year-Old Riemann Hypothesis
The Riemann hypothesis is a 167-year-old puzzle about hidden order in the prime numbers. Anthropic mathematician Levent Alpöge says he and Claude wrote its proof in Lean, a language where the computer checks every step, much like a compiler. At nearly the same time, Stanford postdoc Alessandro Mazullo posted a proof of a generalized version, written over 12 years without AI. These are still claims, but most mathematicians trust formal Lean verification.
AI-processed from TechCrunch; edited by Hamidun News
Imagine receiving a million-line program. The tests pass, the compiler is satisfied, but nobody can explain how it works, not even its author. Something like that is now unfolding in mathematics. Levent Alpöge, a mathematician at Anthropic, says that together with Claude he found a proof of the Riemann hypothesis. It is one of the seven Millennium Prize Problems, and it is 167 years old: it was formulated in 1859. The announcement came in the same week that Anthropic became the first public company among the leading AI labs. "The timing is terrible," Alpöge wrote in an essay.
What the hypothesis is and what has been claimed
Prime numbers are those divisible only by one and themselves: 2, 3, 5, 7, 11. On the number line they look randomly scattered. The Riemann hypothesis proposes that a strict structure hides behind them. If the claimed result holds, it would be one of the most significant results in mathematics in recent decades.
The key detail: Alpöge says he has a solution that compiles in Lean. Lean is an open language in which proofs are written so formally that a computer checks every step itself. For a developer, this is the equivalent of a compiler or a strict test suite: if the file builds, there are no logical gaps. That is why, as TechCrunch reports, most mathematicians are confident the formalized result will hold up. Still, the community has a long road ahead to understand the argument itself.
Alpöge worked with Claude. He describes the process as a "stream of consciousness" through "incredible confusion," and says the model "flew, danced and sparkled" near the finish line. At the same time, he admits to feeling guilty: there is not yet an explanation of the proof that a human can understand. But he has been obsessed with this problem since childhood and wrote that he "could not live with" waiting. Now he promises to take the work to an explanation that people can read.
How colleagues are reacting
Mathematicians are discussing the result in a Zulip chat. Number theorist Chris Birbeck of the University of East Anglia broke down one of the links: the cohomology estimate, he said, follows from an automorphy lifting argument, which relies on a long list of heavy tools from number theory. At least twenty people reacted to that message with the 🤯 emoji. The TechCrunch author candidly admits they got no further than a calculus course and cannot decode it. For us the takeaway is simple: the proof draws on many fields, and checking it by hand is hard.
Alpöge and NYU mathematician Tristan Bakhmatyan also posted another important result related to the Navier–Stokes equations. These describe fluid flow. Exactly this kind of problem kicked off this summer's race between leading labs and mathematicians over unsolved problems.
The person without AI
In those same days, the president of the American Mathematical Society, Stanford professor Ravi Vakil, published a proof of the generalized Riemann hypothesis. It was written by Stanford postdoc Alessandro Mazullo. Vakil stressed that the work is "entirely human, without AI." Mazullo had been invited to Stanford for a year to polish the text and announce the result after it was checked by enough colleagues. But now they decided it was better to post the current draft.
Mazullo worked on the proof for 12 years. In a brief introduction he addressed mathematicians who were dismayed by the pace and tone of the current race: "We are still here. We are very much alive. We are not going anywhere."
What practitioners should take from this
A quick caveat: for now these are claims, not confirmed facts. Lean verification is reliable, but people still have to understand what has actually been proved. Mazullo's draft has not undergone full peer review at all.
For developers and product managers, the takeaway is twofold. First: the pairing of "model plus automatic verifier" works as an honest filter. The model can make mistakes, but Lean does not take its word for anything. The same principle should be applied to code and agents wherever results can be checked automatically: with tests, types, and schemas. Second: the gap between what AI finds and what people can explain is widening. A result that is correct but incomprehensible is hard to use, let alone build on. Mazullo's human-written proof provides exactly what the machine-generated one still lacks: a text that can be read and understood.
- Nov 21, 2026A Developer Had Claude Install 900 Apps and Built a Free Mac Cleaner
- Nov 21, 2026Apple Is Building an AI Agent and Making It Harder for Outsiders to Get Access: Permission Now Takes 14 Steps, Not 4
- Nov 21, 2026Qwen3-0.6B Fine-Tuned on a Mac mini: Errors Dropped to 3 out of 30 from 9
- Nov 21, 2026The Programmer Stopped Being an AI Agent Dispatcher: Multica Takes Over the Handoff Routine
Key Questions Answered
What the hypothesis is and what has been claimed?
Prime numbers are those divisible only by one and themselves: 2, 3, 5, 7, 11. On the number line they look randomly scattered. The Riemann hypothesis proposes that a strict structure hides behind them. If the claimed result holds, it would be one of the most significant results in mathematics in recent decades. The key detail: Alpöge says he has a solution that compiles in Lean. Lean is an open language in which proofs are written so formally that a computer checks every step itself.
How colleagues are reacting?
Mathematicians are discussing the result in a Zulip chat. Number theorist Chris Birbeck of the University of East Anglia broke down one of the links: the cohomology estimate, he said, follows from an automorphy lifting argument, which relies on a long list of heavy tools from number theory. At least twenty people reacted to that message with the 🤯 emoji. The TechCrunch author candidly admits they got no further than a calculus course and cannot decode it.
What are the key details on The person without AI?
In those same days, the president of the American Mathematical Society, Stanford professor Ravi Vakil, published a proof of the generalized Riemann hypothesis. It was written by Stanford postdoc Alessandro Mazullo. Vakil stressed that the work is "entirely human, without AI." Mazullo had been invited to Stanford for a year to polish the text and announce the result after it was checked by enough colleagues. But now they decided it was better to post the current draft. Mazullo worked on the proof for 12 years.
Test Frontier LLMs & Autonomous Agents in One Workspace
Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.