An AI May Have Solved the Riemann Hypothesis: The Psychology of a 167-Year Obsession
Reviewed by a Psychology Postgraduate (MA)
On Friday 20 November 2026, Levent Alpöge, a mathematician at Anthropic in San Francisco, announced that he believes he and Claude, Anthropic’s large language model, have discovered a proof of the Riemann hypothesis — plus the Generalized Riemann Hypothesis. He posted a 4-page note on his personal website and released the argument as Lean computer code, a 14-part archive — with no preprint, no journal submission, and no independent verification yet. Nature, Scientific American, the Wall Street Journal, and IBM reported the claim the same day. That last part matters most: this is a claim, not a confirmed fact.
The short answer: On 20 November 2026, mathematician Levent Alpöge announced that he and Anthropic’s AI, Claude, may have proved the Riemann hypothesis, a 167-year-old problem about prime numbers. A second, separate claim appeared the same day. Neither claim has been independently verified yet, so the hypothesis remains officially unsolved while experts check the work.
The exact search term “riemann hypothesis” is trending on Google today, 21 November 2026, in the US and Canada. People are searching for what it means — whether a machine has just done something humans could not do for 167 years, and how we are supposed to feel about that.
What is the Riemann hypothesis, and why has it resisted proof for 167 years?
The Riemann hypothesis is a claim about how prime numbers are distributed, proposed in 1859 by German mathematician Bernhard Riemann, who was born 200 years ago, in September 1826. Prime numbers — 2, 3, 5, 7, 11, and onward — refuse to fall into any neat pattern, and Riemann proposed a precise rule for where they fall, a rule that would tidy up a vast amount of mathematics if true. For 167 years, the brightest minds have tried and failed.
It is one of the Clay Mathematics Institute’s seven Millennium Prize Problems, each carrying a $1 million prize. Humans alone had solved one of the seven in 26 years; in the last three months, AI appears to have solved two, an OpenAI model reportedly solving one in September 2026.
What exactly was claimed on 20 November?
Alpöge published no traditional paper: a short 4-page note plus the argument as Lean computer code, a formal, machine-checkable 14-part archive. Code can be checked line by line, but 14 parts of dense code still take experts time and patience. James Maynard, a Fields Medal winner at Oxford, said, “If this fully checks out, it is clearly a historic milestone in mathematics.” That “if” is doing a lot of work.
Andrew Granville of the University of Montreal called the released document “garbled” and questioned the timing: the announcement came the day after Anthropic’s record-breaking stock market debut on 19 November, when its shares then fell. Alpöge denies the timing was cynical. Still, when a claim arrives wrapped in a news cycle, the mind reaches for motives before mathematics.
Why did a second claim appear the same day?
The same day, a second, separate claim appeared: a draft by Alessandro Masullo, shared by Ravi Vakil, president of the American Mathematical Society, claiming the Generalized Riemann Hypothesis by an entirely human route — worked on for twelve years and vetted for months by Brian Conrad at Stanford. Two claims, one day, two different routes: one leans on an AI collaborator, the other on twelve years of slow human work. For now, the honest position is uncomfortable: we do not know yet.
Pause and ask yourself: When did you last believe a big claim before it was checked — and what made you trust it: the evidence, the person telling you, or how much you wanted it to be true?
Why do unsolved problems grip the human mind?
Unsolved problems grip us because the mind hates unfinished business. This is the Zeigarnik effect: tasks left incomplete stay active in memory far longer than tasks we finish. A half-read book nags at you; an unanswered text sits in the back of your mind all afternoon. Now scale that up to 167 years: the ultimate unfinished task, a question posed before electric light and carried from one generation to the next. The mind keeps returning to what it cannot resolve.
Picture one familiar evening: you give up on a crossword clue at bedtime and wake with the answer, as if your mind kept working overnight.
Overthinking lives in the same neighborhood. The same loop that keeps a great problem alive can keep a small worry spinning — the email you sent, the thing you said at dinner. The difference is scale, and whether it pulls you toward something meaningful or wears you down.
Why does the end of a mystery feel like grief?
Alpöge himself said he is presenting the argument before having fully mastered the mathematics around it; his statement was described as elegiac, recognizing grief among mathematicians at a mystery ending. Elegiac means mournful, like a lament — a man announcing what could be the greatest result of his life sounded, in part, sad.
That sadness makes sense: a long mystery is a shared project, and solving it ends the shared not-knowing, the long companionship of the search. Choking under pressure and the winner mindset touch the same truth: the pursuit itself shapes us, not only the outcome. Athletes miss the chase after a championship is won.
What is the difference between proof and understanding?
A proof tells you something is true; understanding tells you why. The gap between them is where much of the current anxiety lives. A computer-checked proof can be correct without being understood: every step can check out while no human mind holds the whole argument at once. Terence Tao warned in a September interview that AI could produce solutions faster than mathematicians can understand them. That is the fear: a future in which we know more and understand less.
Can we trust an AI proof we cannot follow?
Trust is the hinge here, and it is psychological before it is technical: we trust a proof when we trust the people and the process behind it. Formal code can be checked mechanically, but somebody still has to do the checking.
The Anthropic IPO shows how quickly trust gets tangled with motive: a claim announced the day after a record-breaking stock market debut invites a question that has nothing to do with mathematics — why now? Alpöge denies the timing was cynical. But the human mind is a motive-detecting machine: we ask who benefits, and whether the announcement is a result or a performance. Our attention makes this harder: “AI solves 167-year-old problem” travels far faster than “unverified claim awaits expert checking.” Gratitude is the counterweight: slowing down to notice what is actually there, not what the headline promises.
How do you live with verification anxiety?
Waiting for experts to check a claim you care about is uncomfortable — call it verification anxiety, the itch to know now. It shows up far beyond mathematics: test results, a job decision, news about someone you love. A few steady habits help:
- Name the status out loud. Say “claimed, not confirmed” before you share the news.
- Separate the claim from the meaning. The proof may check out or it may not. Either way, your reasons for caring — curiosity, wonder, love of a mystery — are already yours.
- Watch one reliable source, not the whole feed. Pick one careful reporter and check back occasionally, not all day.
- Let the experts work. Verification is slow because it is thorough — a quick verdict would be worse.
Frequently asked questions
Has the Riemann hypothesis been solved? No — not yet, and possibly not at all. A proof has been claimed, not confirmed, so the hypothesis remains officially unsolved until experts verify it.
Who claimed the proof, and how? Levent Alpöge of Anthropic announced on 20 November 2026 that he and Claude, Anthropic’s AI, may have proved it, releasing a 4-page note and 14-part Lean code archive. A second claim by Alessandro Masullo appeared the same day.
What happens next? Independent mathematicians will examine the Lean code over weeks or months. If it checks out, it is a historic milestone; if not, the mystery continues.
Try it on our site
A 167-year mystery ending — or not ending — is a lot to sit with. These tools can help:
- Mind Check-In — check in with how you feel.
- Focus Room — a calm space to concentrate amid the headlines.
- Thought Reframe — look at an anxious thought more steadily.
- Brain Games — a different puzzle to chew on while you wait.
Related reading: Anthropic IPO Psychology · How to Stop Overthinking Everything · Choking Under Pressure
Key Takeaways
- On 20 November 2026, Levent Alpöge announced that he and Anthropic’s AI, Claude, may have proved the Riemann hypothesis — unverified, with no preprint.
- A second claim by Alessandro Masullo appeared the same day, after twelve years of human work.
- Proposed by Bernhard Riemann in 1859, it concerns prime numbers and is one of seven $1 million Millennium Prize Problems.
- The Zeigarnik effect explains why a 167-year-old open question grips the mind: unfinished business stays active in memory.
- Proof and understanding differ — a result can check out without any human fully understanding why.
- For now, wait: name the claim as unverified, let experts check, and keep the wonder either way.
This article is for education only and is not a substitute for professional advice. If you are struggling, please consider speaking to a licensed professional in your area.