What every source agrees on
Levent Alpöge, a mathematician at the AI company Anthropic, wrote in a four-page note on Friday that he and the firm's Claude chatbot had proved the Riemann hypothesis, the 1859 conjecture about the distribution of prime numbers Scientific American20 November 2026Open the article See it in the source listNew Scientist20 November 2026Open the article See it in the source listThe Hindu21 November 2026Open the article See it in the source list. The note presents no paper, only a version of the proof written in Lean, a programming language that checks mathematical arguments by computer, released as 14 compressed archive parts New Scientist20 November 2026Open the article See it in the source listMLQ.ai21 November 2026Open the article See it in the source listOfficeChai21 November 2026Open the article See it in the source list. An informal AI-generated write-up of the argument ran to more than 250,000 pages, which Alpöge called too unreadable to post New Scientist20 November 2026Open the article See it in the source listMLQ.ai21 November 2026Open the article See it in the source list. Independent mathematicians including MIT's Andrew Sutherland have compiled the Lean code, and two separate proof-checking systems have accepted it Scientific American20 November 2026Open the article See it in the source listThe Hindu21 November 2026Open the article See it in the source list. The claim has not been reviewed by outside experts, and the Clay Mathematics Institute still lists the hypothesis as unsolved The Hindu21 November 2026Open the article See it in the source listKingy.ai21 November 2026Open the article See it in the source list.
What the sources contradict each other on
The numbers are the sources this piece is built from, listed in full at the foot of the page. Tap one to jump to it.
What the compiling establishes is the first dispute.
One checking report says the Lean code compiles and that two separate proof systems accepted it The Hindu21 November 2026Open the article See it in the source listMLQ.ai21 November 2026Open the article See it in the source list, while the statement-match and comparator runs are still pending, no expert has reviewed the argument, and the Clay Institute still lists the problem as unsolved New Scientist20 November 2026Open the article See it in the source listBBC News21 November 2026Open the article See it in the source listThe Hindu21 November 2026Open the article See it in the source list.
Asserted by MLQ.ai21 November 2026Open the article See it in the source list Disputed by New Scientist20 November 2026Open the article See it in the source listBBC News21 November 2026Open the article See it in the source listThe Hindu21 November 2026Open the article See it in the source list
Who deserves the credit is the second dispute.
The note credits Claude with outpacing the author near the end and with key technical steps MLQ.ai21 November 2026Open the article See it in the source listOfficeChai21 November 2026Open the article See it in the source list, while Alpöge says he guided the work to the right situation, and acknowledges he has not yet mastered all of the mathematics involved The Hindu21 November 2026Open the article See it in the source listMLQ.ai21 November 2026Open the article See it in the source list.
Asserted by MLQ.ai21 November 2026Open the article See it in the source listOfficeChai21 November 2026Open the article See it in the source list Disputed by The Hindu21 November 2026Open the article See it in the source list
An earlier credit fight shadows this week's claim.
New York University mathematician Tristan Buckmaster implied that OpenAI had stolen his and Alpöge's earlier fluid-problem work, one account recalls Scientific American20 November 2026Open the article See it in the source list, while OpenAI denied wrongdoing in that dispute The Hindu21 November 2026Open the article See it in the source list.
Asserted by Scientific American20 November 2026Open the article See it in the source list Disputed by The Hindu21 November 2026Open the article See it in the source list
How each source tells the same story
a sandboxed build on the LUMI supercomputer compiled all 175,187 modules and reproduced the reported hashes
MIT mathematician Andrew Sutherland put the chance that anything is fishy in Alpöge's Lean proof at approximately zero, after independently compiling it Scientific American20 November 2026Open the article See it in the source listMLQ.ai21 November 2026Open the article See it in the source list. The verification runs are unusually public: an independent project compiled five separately written theorem statements against Lean's mathematics library, a second proof checker accepted the full export, and a sandboxed build on the LUMI supercomputer compiled all 175,187 modules and reproduced the reported hashes Kingy.ai21 November 2026Open the article See it in the source list. What remains open is the other half of the bargain: whether the code proves what it says it proves. Buzzard says the first thing to check is whether the Lean statement matches the actual theorem, and Sutherland's public report still lists the end-to-end comparator run, which tests that the code does not redefine symbols or exploit bugs, as pending MLQ.ai21 November 2026Open the article See it in the source listKingy.ai21 November 2026Open the article See it in the source list. The Clay Mathematics Institute still lists the hypothesis as unsolved Kingy.ai21 November 2026Open the article See it in the source list.
Drawn from Scientific American20 November 2026Open the article See it in the source listMLQ.ai21 November 2026Open the article See it in the source listKingy.ai21 November 2026Open the article See it in the source list
the first ten trillion
Bernhard Riemann proposed the hypothesis in 1859: that the nontrivial zeros of a function now called the Riemann zeta function all lie on a single vertical line in the complex plane, which would mean the primes are distributed about as regularly as they could be Tom's Hardware20 November 2026Open the article See it in the source listOfficeChai21 November 2026Open the article See it in the source list. The difficulty is not counting primes but knowing how far any estimate can miss; the count of zeros checked so far runs to the first ten trillion Tom's Hardware20 November 2026Open the article See it in the source listKingy.ai21 November 2026Open the article See it in the source list. Proving the hypothesis would put a square-root-scale ceiling on the error in the prime count, in the standard notation O(sqrt(x) log x) Kingy.ai21 November 2026Open the article See it in the source list. Results across number theory, thousands of them, currently rest on the assumption that it holds OfficeChai21 November 2026Open the article See it in the source list, so the claim reaches the whole field at once. Alpöge says his method also resolves the Generalized Riemann Hypothesis, the same question for a family of functions called Dirichlet L-functions, with an extended version still future work Scientific American20 November 2026Open the article See it in the source listMLQ.ai21 November 2026Open the article See it in the source listKingy.ai21 November 2026Open the article See it in the source list.
Drawn from Scientific American20 November 2026Open the article See it in the source listTom's Hardware20 November 2026Open the article See it in the source listMLQ.ai21 November 2026Open the article See it in the source listOfficeChai21 November 2026Open the article See it in the source listKingy.ai21 November 2026Open the article See it in the source list
about six weeks of work using a non-public version of Claude
The note is generous about who did the late work: "Claude far outpaced me near the finish line," Alpöge writes, after he guided the work to the right situation MLQ.ai21 November 2026Open the article See it in the source listOfficeChai21 November 2026Open the article See it in the source list. A timeline the note says Claude compiled dates the first steps to October 6 and the final step, the Generalized Riemann Hypothesis, to November 16 MLQ.ai21 November 2026Open the article See it in the source listOfficeChai21 November 2026Open the article See it in the source list, after about six weeks of work using a non-public version of Claude BBC News21 November 2026Open the article See it in the source listOfficeChai21 November 2026Open the article See it in the source list. The method adapts Stepanov's technique from function fields, where the analogue of the hypothesis is proved, and Claude is credited with suggesting a long chain of theta lifts to supply the missing multiplicity, each lift adding one unit, the chain roughly as long as the square root of the primes in question MLQ.ai21 November 2026Open the article See it in the source listOfficeChai21 November 2026Open the article See it in the source list. Alpöge is candid: he says he has little mastery of the higher-rank forms where Claude did the technical work, and that he may have baked strange choices into the argument MLQ.ai21 November 2026Open the article See it in the source listOfficeChai21 November 2026Open the article See it in the source list.
Drawn from BBC News21 November 2026Open the article See it in the source listMLQ.ai21 November 2026Open the article See it in the source listOfficeChai21 November 2026Open the article See it in the source list
"an obvious cynical view, to do with the timing, but it is false"
Anthropic listed on the stock market on November 19, and the note appeared the next day New Scientist20 November 2026Open the article See it in the source listBBC News21 November 2026Open the article See it in the source list. Alpöge had messaged chief executive Dario Amodei on the Sunday before, asking how to announce the discovery, and says he determined there was a constraint to respect before the announcement could be made New Scientist20 November 2026Open the article See it in the source listBBC News21 November 2026Open the article See it in the source list. He does not pretend the optics are good: "The precise timing is horrible," he writes, adding that if he could have, he would have finished it all much earlier BBC News21 November 2026Open the article See it in the source list. To the cynics he answers on Friday with words of his own: "an obvious cynical view, to do with the timing, but it is false", he writes BBC News21 November 2026Open the article See it in the source listNew Scientist20 November 2026Open the article See it in the source list.
Drawn from New Scientist20 November 2026Open the article See it in the source listBBC News21 November 2026Open the article See it in the source list
both a wonderful and a terrible time to be a research mathematician
Brian Conrey, a leading specialist on the hypothesis, called it a monumental moment in human history and really the biggest thing that has ever happened in mathematics, while noting he had not yet seen a manuscript and that it looks seriously complicated The Daily Star21 November 2026Open the article See it in the source list. Hector Pasten of the Pontifical Catholic University of Chile says a correct proof completes a dictionary between different branches of mathematics, something like a Rosetta Stone Scientific American20 November 2026Open the article See it in the source list. Carlo Pagano of Concordia says the proof would open brand-new avenues of thinking in number theory Scientific American20 November 2026Open the article See it in the source list. Royal Holloway's James McKee notes that numerous results conditional on the generalized hypothesis would become theorems, and calls it both a wonderful and a terrible time to be a research mathematician New Scientist20 November 2026Open the article See it in the source list. Sutherland, who once said an AI that could prove the Riemann Hypothesis was not one he would want to meet, now calls the day a human triumph and a day for celebration Scientific American20 November 2026Open the article See it in the source list.
Drawn from Scientific American20 November 2026Open the article See it in the source listNew Scientist20 November 2026Open the article See it in the source listThe Daily Star21 November 2026Open the article See it in the source list
Where the sources part company
What the compiling establishes divides the accounts. Sutherland puts the chance that anything is fishy at approximately zero, and says the kernel rechecked the proof on three standard axioms Scientific American20 November 2026Open the article See it in the source listMLQ.ai21 November 2026Open the article See it in the source listKingy.ai21 November 2026Open the article See it in the source list; Imperial College's Kevin Buzzard says the first question is whether the Lean statement matches the theorem, a test still pending; and Tao, a prominent mathematician, says the digestion will take some time New Scientist20 November 2026Open the article See it in the source listBBC News21 November 2026Open the article See it in the source listMLQ.ai21 November 2026Open the article See it in the source list. Conrey, a leading specialist on the hypothesis, calls it the biggest thing that has ever happened in mathematics The Daily Star21 November 2026Open the article See it in the source list, Royal Holloway's James McKee calls it both a wonderful and a terrible time to be a research mathematician New Scientist20 November 2026Open the article See it in the source list, and a group of maths scholars has said AI companies' habit of boasting of major solutions is "detrimental to the science of mathematics" BBC News21 November 2026Open the article See it in the source list. The note appeared a day after Anthropic's market debut, and Alpöge says the obvious cynical reading is false Scientific American20 November 2026Open the article See it in the source listNew Scientist20 November 2026Open the article See it in the source listBBC News21 November 2026Open the article See it in the source list.
Sources
- Scientific American0120 November 2026
- New Scientist0220 November 2026
- BBC News0321 November 2026
- The Hindu0421 November 20260521 November 2026
- The Daily Star0621 November 2026
- Tom's Hardware0720 November 2026
- MLQ.ai0821 November 2026
- OfficeChai0921 November 2026
- Kingy.ai1021 November 2026
The small print
This article synthesises publicly available reporting for informational purposes. The perspectives summarised are paraphrased from the listed publications and reflect their respective editorial framing. Plural does not independently verify, endorse, or adopt the underlying factual claims, and provides this information on an 'as-is' basis without assuming liability for the accuracy of the original sources or the content of external links.
This summary represents an attempt at neutrality and may still contain bias; it is not presented as absolute truth.
No story runs here unless at least six genuinely independent newsrooms have covered it, counted after copies of the same wire report are collapsed into one. AI reads those accounts and writes the draft from them, and may work from nothing else. Fifty-nine automated checks then run on what it wrote: every name, figure, date and quotation must be found in the accounts the draft cites, and a draft that adds an outside detail, infers a motive or expresses an opinion is rejected. Thirty-nine of those checks block publication outright, and a failing draft is discarded, never salvaged.
What survives this process is merely a draft, not a published piece. A human editor then scrutinises the text line by line against the original sources to minimise bias and verify accuracy. No content reaches a reader any other way.