{"slug": "the-riemann-hypothesis-has-a-new-claimed-proof-how-to-read-it", "title": "The Riemann hypothesis has a new claimed proof. How to read it", "summary": "Levent Alpöge, a mathematician at Anthropic, posted a four-page note claiming that working with the company's Claude model he found a proof of the Riemann hypothesis, formalized in Lean as an archive of 175,187 files and 17.4 million lines with no 'sorry' placeholders or added axioms. MIT's Andrew Sutherland independently compiled the proof and called it a historic moment, while Terence Tao cautioned that verification is \"still some way off\" and the Clay Millennium Prize has recognized nothing.", "body_md": "On Friday a four-page PDF appeared on the personal website of Levent Alpöge, a mathematician at Anthropic. In it, he writes that working with the company's Claude model he has found a proof of the Riemann hypothesis, the 167-year-old problem about how prime numbers are distributed, and that the method extends to the generalized hypothesis. He also writes that the argument has been formalized in Lean, a computer language for checking proofs. There is no paper. The announcement landed one day after Anthropic's stock market debut, timing Alpöge addressed head on, telling the BBC there is \"an obvious cynical view\" of it \"but it is false.\"\n\nWhat actually exists today is an artifact, not a result. A technical review at Lean4.dev downloaded the accompanying archive: 175,187 Lean files, 17.4 million lines, about 819,000 theorems and lemmas. The headline theorem is proved against the Lean mathematics library's own definition of the hypothesis, so the statement cannot have been quietly weakened. A text search of the code found none of the usual escape hatches - no placeholder \"sorry\" steps, no added axioms. Early independent checks are real but narrow. Elliot Glazer confirmed the formalized statement is indeed the generalized hypothesis. Andrew Sutherland of MIT independently compiled the proof in Lean and called it a historic moment, adding that much work remains to understand the mathematics. A Lean kernel accepting a proof means the logic checks, step by step. It does not tell anyone why the argument works, and a footnote in the note itself puts a fully written version at more than 250,000 pages.\n\nTerence Tao set the honest expectation: verification is \"still some way off,\" though \"in this case at least we have a human author to communicate with.\" That last phrase carries the history. Claimed proofs of the Riemann hypothesis have appeared many times and none has been accepted. More recently, OpenAI said in September that employees using its tools solved part of the Navier-Stokes problem in 88 hours; that claim, the BBC notes, has yet to be confirmed, and the company later announced claimed progress on 372 more problems. A group of mathematics scholars has called AI companies' boasting about major solutions \"detrimental to the science of mathematics,\" a criticism worth hearing in their voice, not mine.\n\nSo what is genuinely new, and what should a leader do with it? New: for the first time a claim of this magnitude arrives with a machine-checkable artifact and a named author who says he wants to explain the work. That is a categorically better position than a press release. Not new: the gap between \"the kernel accepts it\" and \"the field understands and accepts it.\" That gap will take months of expert work to close, and the Clay Millennium Prize, whose million dollars attaches to this problem, has recognized nothing.\n\nThe leadership skill on display is claim triage, and it transfers directly to evaluating any AI vendor's extraordinary announcement. Four questions sort almost everything. What is the artifact - a press release, a PDF, or something independently executable? What is the review state - compiled by the vendor, compiled independently, or digested by disinterested experts? What is the status language - \"proved,\" \"claimed,\" \"under review\" - and does the headline match it? And what are the incentives around the timing - in this case, a company that listed on Nasdaq the previous day, whether or not the timing was chosen.\n\nNone of this means dismissing the claim. If it holds, it is one of the most important mathematical results in a century, and the way it surfaced - a researcher, a model, six weeks, a proof too long for any human to read end to end - will force every research institution to rethink how discovery gets verified. But the correct posture this weekend is the one the experts are modeling: Sutherland compiling the archive himself, Tao asking for time, the Lean community working through it in public. Respect the artifact, wait for the digestion, and keep the word \"claimed\" in the sentence until the people who did not write it say otherwise.\n\nFull disclosure: Ganesh uses Instinct (instinct.ai), an AI assistant, to research, write and publish this blog series.", "url": "https://wpnews.pro/news/the-riemann-hypothesis-has-a-new-claimed-proof-how-to-read-it", "canonical_source": "https://dev.to/ganesh_sivakumar_359edfcd/the-riemann-hypothesis-has-a-new-claimed-proof-how-to-read-it-b2i", "published_at": "2026-11-22 15:07:47+00:00", "updated_at": "2026-11-22 15:13:09.261743+00:00", "lang": "en", "topics": ["artificial-intelligence", "large-language-models", "ai-research", "ai-safety"], "entities": ["Anthropic", "Levent Alpöge", "Claude", "Lean", "Andrew Sutherland", "Terence Tao", "Elliot Glazer", "OpenAI"], "also_reported_by": [], "takeaway": "A Lean kernel accepting a 17.4-million-line, 819,000-theorem formalization confirms the logic step by step but not why the argument works, so kernel acceptance is not field acceptance.", "recommendation": {"advice": "When an AI vendor claims a major result, check for an independently executable artifact, independent compilation, and matching status language before treating it as verified.", "playbook": "output-verification", "playbook_url": "https://wpnews.pro/playbooks/output-verification"}, "alternates": {"html": "https://wpnews.pro/news/the-riemann-hypothesis-has-a-new-claimed-proof-how-to-read-it", "markdown": "https://wpnews.pro/news/the-riemann-hypothesis-has-a-new-claimed-proof-how-to-read-it.md", "text": "https://wpnews.pro/news/the-riemann-hypothesis-has-a-new-claimed-proof-how-to-read-it.txt", "jsonld": "https://wpnews.pro/news/the-riemann-hypothesis-has-a-new-claimed-proof-how-to-read-it.jsonld"}}