From our other sites
The story
A mathematician at the US company Anthropic has announced that the firm’s generative AI found a proof of the Riemann hypothesis, unsolved for 167 years. The proof reportedly runs to over 250,000 pages, with the code used for formal verification totaling 17.4 million lines — a scale that left 5ch users both skeptical (“can a human even peer-review this?”) and worried about the fallout for prime-number-based encryption. Peer review isn’t finished yet, and the thread was further stirred by news that, around the same time, a separate mathematician independently claimed to have proven the generalized Riemann hypothesis. Reactions split between cautious wait-and-see over verification and genuine excitement.
The Riemann Hypothesis, Unsolved for 167 Years, Possibly Proven by Generative AI… Anthropic Mathematician Announces
A mathematician at US AI developer Anthropic announced on his personal website on the 20th that, using the company’s generative AI, he had found a proof of the notoriously hard “Riemann hypothesis.”
If correct, it would resolve a problem mathematicians have grappled with for 167 years.
Source: yomiuri.co.jp / Original article here
What people said
Some people were saying existing methods just couldn't crack it and that some genuinely new mathematics would be needed… so how big a conceptual leap is this, really?
Guess we'll just have to wait for a natural-language writeup of the proof.
Neither one has been peer-reviewed yet, but it sounds like things are about to get hectic on the Riemann hypothesis front for the next few days.
It'd be funny if he just had the AI take a crack at it, went 'oh, THAT'S the trick!,' and then finished the proof by hand himself.
Though I guess if you just get inspired from reading someone else's paper, that doesn't really count as 'getting help.'
Even in applied science, more and more drug discovery has AI designing the proteins and peptides.
The hard part now is probably getting physical-world AI to operate using the vocabulary of someone with, say, a 40 hensachi (a Japanese standardized test score where 50 is average — so, well below-average verbal ability).
Seems like that one's beyond what the human brain can manage.
It's always like this.
Nah, even if this gets proven, it's not like prime-number encryption suddenly becomes invalid.
Apparently encryption has nothing to do with it.
Now THAT, even an AI could pull off for sure.
The New York Times has a story on it too
First go prove the ABC conjecture, lol
And the verification tools themselves can't even guarantee they're bug-free, yet they're just being used as a benchmark anyway.
Apparently it's already passed multiple verification tools. No mathematicians are actually complaining about it.
Let's talk once peer review is actually done.
…Or are they gonna have AI do the peer review too?
According to the source it's apparently over 250,000 pages. And the verification already uses AI.
> over 250,000 pages
That is incredibly sketchy ^^;
Exploring that corner of the mathematical world is simply beyond the human brain.
Thanks for all your hard work over the years, mathematicians.
November 21, 2026
https://www.asahi.com/articles/ASVCP10Z2VCPUHBI00GM.html
November 21, 2026 9:44 AM
https://www.nikkei.com/article/DGXZQOSG20CQC0Q6A121C2000000/
167-Year-Old Unsolved Math Problem
2026/11/21
https://www.sankei.com/article/20261121-ZYCHSZN7TBNLXNXFBDAF4JTLGQ/
"Anthropic mathematician claims to prove Riemann hypothesis, most famous problem in all of math", Scientific American
https://www.scientificamerican.com/article/anthropic-mathematician-claims-to-prove-riemann-hypothesis-most-famous-problem-in-all-of-math/
But realistically, that 'tidying up' is probably an extraordinarily non-trivial task in itself (if it were easy, they'd have already done it).
Background and Key Points of This Topic
The Riemann hypothesis is an unsolved problem proposed in 1859 that’s said to govern the distribution of prime numbers, and it’s one of the Millennium Prize Problems carrying a $1 million reward. Worries that “prime-number encryption will be broken” surfaced in the thread, but as commenters noted, the truth of the hypothesis isn’t directly tied to encryption security — theories have long been built on the assumption that the hypothesis is true anyway, so even a confirmed proof wouldn’t make encryption instantly obsolete. Where opinion really split was over verifiability: with the proof running over 250,000 pages and the AI’s formal-verification code (in Lean) totaling 17.4 million lines, several commenters questioned whether human mathematicians can realistically peer-review something of that scale. At the same time, there are reports it has already passed multiple verification tools, so a wariness informed by past cases of AI “proof achieved” claims later falling apart is mixing with real excitement over a potential breakthrough. One key premise worth noting from the article itself is that this is a pre-peer-review announcement — it is not yet a confirmed result.
*This article is compiled from excerpts and a summary of the 5ch (Science News+) thread “[Mathematics] The “Riemann Hypothesis,” Unsolved for 167 Years, Possibly Proven by Generative AI… Anthropic Mathematician Announces.”
Leave a Reply