ExitsIPOs, listings & liquidity
Anthropic mathematician posts Lean proof of the Riemann hypothesis built with Claude as Anthropic debuts on the stock market
Levent Alpöge, a mathematician at Anthropic, released Lean code that he says proves the Riemann hypothesis and its generalized form, worked out with Claude. An MIT researcher independently compiled the code, but experts have not yet accepted the proof.
From the TBPN podcast episode of November 21, 2026 and cited sources: Alpöge posted a short statement and a large Lean formalization that he says proves both the Riemann hypothesis and the generalized Riemann hypothesis (GRH). [3] [4] [10]
SOURCES[3] Yahoo/science: proof almost certainly true yahoo.com[4] Phys.org phys.org[10] Nacionale: not verified yet nacionale.com
AI-assisted, not human-edited. Methodology Corrections Who owns this site
- What happened
- Alpöge posted a short statement and a large Lean formalization that he says proves both the Riemann hypothesis and the generalized Riemann hypothesis (GRH). [3] [4] [10]
- Original source
- TBPN podcast episode, independent perspective
- Brief by
- AT Pulse research desk AI-assisted; publisher and contact on the desk page
- Studies
- arXiv 1808.10774 (Aug 2018) [12] date from arXiv ID
- When
- November 21, 2026 podcast episode published
- Covered
- Nov 21 edition
- Evidence
- 13 sources across 11 sites; no transcript
Reach: not measured yet

Alpöge posted a short statement and a large Lean formalization that he says proves both the Riemann hypothesis and the generalized Riemann hypothesis (GRH). He described reaching the result through repeated exchanges with Claude. The reviewed sources do not say which Claude version he used. MIT's Andrew Sutherland independently recompiled the code against Lean's kernel using only the three standard axioms, and he said the risk of an error was close to zero. [3] [4] [10]
A clean Lean compile shows that every step checks. It does not settle everything. Kevin Buzzard notes that reviewers still have to confirm that the formal theorem statement actually matches the Riemann hypothesis. Reviewers were also waiting for a comparator run to rule out hidden redefinitions of terms or an exploited kernel bug. Several people have reported issues that must be resolved, and the Clay Institute requires peer-reviewed publication plus two years of scrutiny before it accepts a proof. [4] [10] [11]
The claim came alongside Anthropic's market debut. The IPO priced at $83 a share, raised more than $81 billion and closed up about 17% on its first day. On the next trading day the S&P 500 fell 3.70% and the Nasdaq 5.28%. Nvidia dropped 14%, and Anthropic shares fell 8%, nearly erasing their first-day gain but staying just above the IPO price. [6] [9] [13]
an independent Lean recompilation and a pending expert review. Implication: if the proof holds, formal verification becomes the practical way to check AI-generated reasoning.
Read the full assessment
Investors, however, treated the claim as background to a broader sell-off in AI stocks.
The reviewed sources disagree on some details. I'm writing up the dossier now.
Executive brief
On November 20, 2026, Levent Alpöge, a mathematician at Anthropic, said he and Claude had proved the Riemann hypothesis. He posted the proof as computer-checkable Lean code. MIT's Andrew Sutherland compiled it independently and said the chance of a problem was "approximately zero" (Yahoo/science). The proof is still not accepted: experts have not finished reviewing it, others have reported issues, and the Clay Institute requires peer-reviewed publication plus two years of scrutiny (Phys.org). The claim came one day after Anthropic's record IPO, as a sell-off in AI stocks pushed the S&P 500 into a correction.
What changed and event timeline
Earlier Claude math claim
Alpöge tweeted that Claude had solved the Jacobian conjecture, an earlier example of him publicly announcing AI-assisted results ().
Anthropic lists at $83
The IPO raised more than $81 billion, the largest in U.S. history. Shares closed up about 17%, while the Nasdaq fell 2.25% ().
Riemann claim posted
Alpöge released a short statement and Lean code covering both the Riemann hypothesis and the generalized version (GRH). He said a full written argument would be extremely long (). TheJournal.ie dates the announcement November 21 ().
Rival Stanford proof
Stanford professor Ravi Vakil said a postdoc in his department had independently submitted a proof. BigGo names him as Alessandro Masullo and reports the work as "all human" ().
Market correction
The S&P 500 fell 3.70% and the Nasdaq 5.28%. Nvidia dropped 14%, and ANT fell 8% but stayed narrowly above its IPO price ().
Capabilities and access
- Model: none of the reviewed sources say which Claude version was used (Phys.org).
- Artifacts: a short statement plus a large body of Lean code, posted publicly (search summary of Nacionale).
- Anthropic: no official company statement is quoted. The claim appears to be Alpöge's personal announcement (Scientific American).
Read the full section
- Model: none of the reviewed sources say which Claude version was used (Phys.org).
- Artifacts: a short statement plus a large body of Lean code, posted publicly (search summary of Nacionale).
- Anthropic: no official company statement is quoted. The claim appears to be Alpöge's personal announcement (Scientific American).
- Process: Alpöge describes reaching the result in mid-November through repeated back-and-forth with Claude. He says he has not fully mastered all of the mathematics involved.
Technical analysis for researchers and developers
- What Lean checks: Lean is software that mechanically verifies each step of a proof.
- What Lean does not check: Kevin Buzzard (Imperial College London) points out that the formal theorem statement still has to be confirmed to match the actual Riemann hypothesis.
- Statement scope: the theorem in the code states GRH, which implies the ordinary hypothesis (search summary).
Read the full section
- What Lean checks: Lean is software that mechanically verifies each step of a proof. Sutherland reports that Lean's kernel (its core checker) independently rechecked the proof using only the three standard axioms (Yahoo/science).
- What Lean does not check: Kevin Buzzard (Imperial College London) points out that the formal theorem statement still has to be confirmed to match the actual Riemann hypothesis. Reviewers were also waiting for a "comparator" run, which tests whether the code quietly redefines terms or exploits an unknown kernel bug.
- Statement scope: the theorem in the code states GRH, which implies the ordinary hypothesis (search summary).
- Implication: a formal check can be reproduced by anyone, but human understanding of the proof lags behind it.
Claims and evidence
- Independent: Sutherland (MIT) rechecked it (Yahoo/science).
- Claimed by the author, an Anthropic employee whose employer had just gone public. Not peer-reviewed (Phys.org).
- Brian Conrey, a Riemann hypothesis specialist, said this after reading only Alpöge's posted remarks, not the full proof (TheJournal.ie).
Read the full section
- Compiles cleanly in Lean. Independent: Sutherland (MIT) rechecked it (Yahoo/science).
- Proves RH and GRH. Claimed by the author, an Anthropic employee whose employer had just gone public. Not peer-reviewed (Phys.org).
- "Monumental." Brian Conrey, a Riemann hypothesis specialist, said this after reading only Alpöge's posted remarks, not the full proof (TheJournal.ie).
- Opens new research directions. Carlo Pagano (Concordia University, visiting researcher at Google DeepMind) called the new ideas "very plausible" routes to further work (Scientific American).
Context and prior work
- AI and math in 2026: Scientific American lists OpenAI's October work on prime numbers and a disputed September claim about the Navier–Stokes equations (Scientific American).
- Failed past proofs: human-written proofs of the Riemann hypothesis have appeared on arXiv for years without being accepted (for example arXiv 1808.10774).
- Status: the Clay Institute still lists the problem as unsolved (search summary).
Read the full section
- AI and math in 2026: Scientific American lists OpenAI's October work on prime numbers and a disputed September claim about the Navier–Stokes equations (Scientific American).
- Failed past proofs: human-written proofs of the Riemann hypothesis have appeared on arXiv for years without being accepted (for example arXiv 1808.10774).
- Status: the Clay Institute still lists the problem as unsolved (search summary).
Limitations, safety and contested findings
- Reported issues: several people have reported problems in the proof, some found with AI help. All of them must be resolved before acceptance (search summary).
- Rival proof: sources disagree on whether the Stanford postdoc's proof was human-only (BigGo) or Claude-assisted in Lean (search summary of the same coverage).
- Timing: TheJournal.ie notes the announcement came right as Anthropic debuted on the stock market.
Read the full section
- Reported issues: several people have reported problems in the proof, some found with AI help. All of them must be resolved before acceptance (search summary).
- Rival proof: sources disagree on whether the Stanford postdoc's proof was human-only (BigGo) or Claude-assisted in Lean (search summary of the same coverage). The TheJournal.ie report does not say either way.
- Timing: TheJournal.ie notes the announcement came right as Anthropic debuted on the stock market.
- Author's understanding: Alpöge admits he does not fully understand every part of the proof.
Business and practitioner implications
- Investors: the reported IPO figures ($83 price, more than $81 billion raised) come from news reports.
- Macro worries: Dan Niles reports combined OpenAI and Anthropic sequential revenue growth slowing from 75% to 48% while losses rose (CNBC).
- Developers: formal verification (Lean plus independent checking) is becoming the standard way to prove that AI-generated reasoning is correct.
Read the full section
- Investors: the reported IPO figures ($83 price, more than $81 billion raised) come from news reports. No valuation was confirmed in the reviewed sources (CNBC). The proof claim did not stop the sell-off; Bloomberg reports ANT nearly gave back its first-day gain (Bloomberg).
- Macro worries: Dan Niles reports combined OpenAI and Anthropic sequential revenue growth slowing from 75% to 48% while losses rose (CNBC).
- Developers: formal verification (Lean plus independent checking) is becoming the standard way to prove that AI-generated reasoning is correct.
The source trail.
Sources (13)
AI Solves Riemann Hypothesis, Stock Market Plunges, The Mansion Section | Diet TBPN
Metadata only. Transcript unavailable: Audio transcription returned HTTP 429.
share.transistor.fmMarket Nukes, Math Solved Again, Mansion Section | Alex Epstein, David Chang, Zhou Xian, Eric Wind
Related coverage; assess separately
share.transistor.fm