AdvertiseFee Deal Desk

Macro

Anthropic's Alpöge Claims Riemann Proof of 167-Year-Old Puzzle

Anthropic mathematician Levent Alpöge says Claude helped prove the Riemann hypothesis, releasing a Lean proof archive one day after Anthropic's Nasdaq IPO.

Be a creator
November 20, 2026, 09:42 PM UTC4 min read
AI SummaryAI
  • Anthropic mathematician Levent Alpöge claims Claude helped prove the Riemann hypothesis.
  • Alpöge published the claim on November 20, one day after the IPO.
  • The Lean proof archive is split into 14 downloadable parts.
  • The Clay Mathematics Institute attaches a $1 million prize to the hypothesis.

The Claim and the IPO Timing

Levent Alpöge, a mathematician at Anthropic, says the company's Claude model helped him construct a proof of the Riemann hypothesis, the 167-year-old conjecture that carries a $1 million Millennium Prize from the Clay Mathematics Institute. The claim surfaced on Friday, November 20, one day after the company's stock-market debut on Nasdaq, which followed an offering that raised $81.34 billion. No independent verification of the result exists yet, and the announcement's position on the calendar, immediately after a debut that raised tens of billions, has itself become part of the story.

The four-page note states that a proof of the Riemann Hypothesis was found with Claude's assistance and that the same method settles the broader Generalized Riemann Hypothesis, which extends the question to other mathematical functions. Alpöge encoded the argument in Lean, a proof assistant that checks formal deductions step by step, and his note links to an archive split into 14 downloadable parts. Availability alone does not move the mathematics: the community still has to reproduce the archive, confirm the formalized theorem matches the original conjecture and examine the claimed result before treating it as settled.

The timing has an internal record of its own. Alpöge says he messaged Anthropic CEO Dario Amodei about a possible announcement late on Sunday, November 15, then consulted company leadership and was told there was a constraint to respect, which he did not describe further. He called the precise timing horrible, while rejecting a cynical reading of why the claim appeared when it did and stopping short of blaming the IPO. Under the institute's prize rules, a proposed solution must be published in a qualifying outlet, survive at least two years of scrutiny and earn general acceptance before any award can be considered. Anthropic had previously reported progress on the zeta function without claiming the full hypothesis; the new assertion goes far beyond that earlier work.

From Stepanov Determinants to Theta Lifts

His public research log lays out how the work got here. Exploration with Claude began by October 6 around Stepanov determinants, a tool from a branch of number theory; mid-October brought in automorphic forms and the relative trace formula; and the work pivoted on October 20 to sums built from the von Mangoldt function. The decisive entries cluster on November 15, covering the construction of continuous theta lifts, norm computations along the lift, divisibility bounds and links between the structures involved. The log's final entry, dated November 16, is marked simply GRH, the abbreviation for the generalized hypothesis.

The core idea of the late stage is Claude's, as the note tells it. Techniques proven on function fields resist direct transfer to the integer setting, and the model proposed running chains of consecutive theta lifts to simulate multiplicity, with a lift chain of length roughly √X for a von Mangoldt sum of length X, alongside the Stepanov method and Fourier analysis. Alpöge writes that near the finish line Claude was advancing at a pace he struggled to match and that its command of areas such as automorphic forms on high-rank groups exceeded his own, leaving him holding a complete formal proof he still needs time to fully understand. An attempt to render the argument in human-readable form ballooned past 250,000 pages, so the Lean archive went out first, with a formal paper to follow.

The claim lands in a field already on edge. OpenAI claimed in September that it had solved the Navier-Stokes Millennium Prize problem while Alpöge and NYU mathematician Tristan Buckmaster pursued related fluid-equation work, then followed in October with 722 manuscripts from an unreleased model, grouped into 372 result families, only 162 of them carrying Lean-checked main results; mathematicians demanded stronger evidence rather than taking the batch at face value. Toronto mathematician Daniel Litt has warned that, without adaptation, there would be no more math in 50 years, and Buckmaster argued at an MIT panel this week that mathematics still requires human understanding.

Verification, Not the Announcement, Decides

Both events this week, the claim and the debut, funnel into the same bottleneck: whether anyone outside Anthropic can reproduce the 14-part archive and check that its formalized theorem actually states the 1859 conjecture. October's 722-manuscript release set the community's reference point, and its lesson was that a flood of claims without reproduction convinced no one. Clay's rules guarantee the $1 million question stays open for at least two years even if every check passes, so the near-term test is narrower: independent mathematicians rebuilding the archive, and Alpöge delivering the readable paper he has promised. The one-day gap to the IPO will stay part of the debate regardless of how the mathematics resolves.

COINOTAG's editorial and research desk.

AI-Assisted

AI-generated, AI-reviewed, under COINOTAG editorial oversight.

Price-Impacting News

Live
All breaking news
Breaking News