Tech • AI • Robotics • Game

ENFR

Full article — scored 10/10

Anthropic Researcher Says Claude Helped Him Prove Riemann Hypothesis

Levent Alpöge’s claim that Claude helped produce a Lean-formalized proof of the Riemann hypothesis has triggered a fast-moving verification effort, a rival human draft, and a deeper question for mathematics: when an AI-assisted proof exists first as code, what does it mean for the field to understand it?

Sign in to follow
Generated November 21, 2026 at 3:44 AM≈ 7 min read
Original source — BigGo Finance

A claim big enough to move slowly

The story began, publicly, with a four-page note on the personal website of Levent Alpöge, a mathematician at Anthropic, saying he believed he had discovered a proof of the Riemann hypothesis while working with Claude, Anthropic’s AI model . The statement, if it survives expert scrutiny, would be one of the largest mathematical events in modern history: the Riemann hypothesis was formulated in 1859 and has shaped number theory for 167 years . Alpöge also claimed that the method extends to the generalized Riemann hypothesis, a broader family of statements about L-functions .

The correct first verb is still “claimed,” not “proved.” Current reporting and expert commentary emphasize that the result is not yet an accepted theorem, because mathematicians must verify both the formal Lean artifact and the match between its formal statement and the standard mathematical claim . State of Proof, which opened an intake record on November 20, lists the matter as unexamined and says the next bounded check is to identify the relevant Lean declarations, compare them with the classical and Dirichlet versions of the hypotheses, inspect the dependency tree and replay the declarations in isolation .

That distinction matters because Lean can check a formal proof only of the statement it is given. If the code proves exactly the intended theorem, with acceptable axioms and no hidden assumptions, the Lean artifact could settle validity faster than a conventional manuscript. If it proves a nearby but weaker or misstated proposition, the Riemann hypothesis remains open, even if the formal work is impressive .

What Alpöge says Claude contributed

Alpöge’s note, as reported by the Washington Post and Quanta, describes a collaboration in which the human researcher supplied mathematical direction and Claude eventually outpaced him near the end of the argument . He did not release a traditional paper; instead, he released a large Lean formalization that he said contains the proof . Quanta reported that the Lean material amounts to about 17 million lines of code, and that Alpöge said attempts to get a human-readable version would have produced more than 250,000 pages .

The mathematical hint in the announcement is not a full proof sketch, but it is specific enough to orient experts. Alpöge’s reported approach descends from Stepanov-style ideas used in the function-field analogue of the Riemann hypothesis, replacing a missing notion of multiplicity over the integers with chains of theta lifts . Terence Tao summarized the public situation by saying Alpöge, using an internal AI model, had generated an “enormous” Lean-only proof of the generalized Riemann hypothesis and that the strategy seemed, from very initial analysis, to use modern Langlands-program and arithmetic-geometry machinery .

For non-specialists, that description means the claim is not that Claude simply guessed a clever elementary trick. It appears, at least from early expert reading, to involve a large synthesis of advanced number theory, automorphic methods and arithmetic geometry . The most important unresolved question is whether this synthesis has been formalized correctly, not whether the headline sounds plausible.

Why Lean changes the verification race

A conventional proof of the Riemann hypothesis would usually be judged through months or years of seminars, private checking, journal review and attempts to simplify the argument. A Lean proof changes the front end of that process. A computer can check whether the formal terms type-check under specified assumptions, and that check can be repeated by independent Lean users if the files, toolchain and dependencies are available .

But formal verification does not eliminate every human question. Mathematicians still need to know whether the formal theorem is the Riemann hypothesis as normally stated, whether the definitions imported from libraries are the right ones, whether the proof depends only on accepted axioms, and whether the community can extract a coherent human strategy from the code . Tao wrote that the Lean discussion was the likely place to follow efforts to verify the proof and extract the strategy, and that the digestion process would take time even with a human author available to answer questions .

That “digestion” is not just public relations. Mathematics values proof as explanation, not only as certification. If a Lean artifact is valid but too large for any person to understand, it may establish truth while leaving the community with a second task: turning formal correctness into conceptual knowledge.

A rival human draft appears the same day

The day did not belong only to Anthropic and Claude. Ravi Vakil, a Stanford professor and president of the American Mathematical Society, published what he called a hurried announcement about Alessandro Masullo’s proposed proof of the generalized Riemann hypothesis . Vakil wrote that Brian Conrad had been looking closely at Masullo’s work for months, had given extensive feedback, and that it was “so far looking good”; he also emphasized that the draft was “all human, with no AI” .

Vakil’s post says the original plan had been to bring Masullo to Stanford for a year to polish the writing and announce later after more people had reviewed it, but that the current climate made it seem best to release the draft immediately . Quanta described the manuscript as 940 pages and said mathematicians were withholding judgment, in part because a previous Riemann-related proof Masullo had worked on years earlier did not ultimately hold up .

The two claims are therefore not the same story in two forms. Alpöge’s is an AI-assisted, Lean-first claim, attributed to work with Claude and released without a conventional explanatory paper . Masullo’s is a human-written draft, released early, with an invitation for checking, formalization and gap-filling . Both concern the generalized Riemann hypothesis, and both remain proposed until the relevant communities finish their work .

The timing problem

Alpöge’s note also acknowledged awkward timing. The Washington Post reported that he said the timing had nothing to do with Anthropic’s recent stock-market debut and that he had consulted Anthropic leadership before announcing . Bloomberg Law likewise reported that the claim came on the heels of Anthropic’s initial public offering and that Alpöge had published a note and shared Lean code as evidence for the claim .

That timing is why caution is essential. A proof of the Riemann hypothesis would be a scientific event, but the surrounding environment is corporate, competitive and emotionally charged. Anthropic builds the model credited in the announcement. Mathematicians are already debating what happens when major open problems are attacked with proprietary systems, large compute budgets and formal artifacts that are hard to read as ordinary mathematics .

The emotional dimension should not be dismissed. The Washington Post quoted Alpöge’s own sense that the problem had been a guiding light and that its possible resolution brought both wonder and loss . For many mathematicians, the Riemann hypothesis has functioned not only as a target but as a generator of methods, questions and careers.

What would count as confirmation?

Near-term confirmation will likely arrive in layers. First, Lean users can attempt to compile and independently check the files. Second, experts can inspect the theorem statements and axioms. Third, number theorists can decide whether the formal proof corresponds to a valid mathematical strategy and whether a human-readable exposition can be written . Only after that would the broader mathematical community begin to treat the result as settled.

The Clay Mathematics Institute’s prize process is separate from online verification. Reporting from AFP via The Straits Times noted that a solution to a Millennium Prize Problem must be published in a peer-reviewed journal and withstand two years of scrutiny to qualify for the $1 million prize . Even a successfully checked Lean proof would therefore not instantly end the formal prize timeline .

The cautious bottom line

As of the latest reports within the November 20–21 window, the current state is clear: Alpöge says Claude helped him prove the Riemann hypothesis and the generalized Riemann hypothesis; a large Lean formalization has been released; independent verification and interpretation are underway; and a separate human draft by Masullo has appeared in parallel . The claim may become a landmark for both mathematics and AI-assisted research, but the responsible position is to treat it as a potentially historic proof claim, not yet as a theorem.

If it holds, the breakthrough would show Claude not merely as a calculator or writing assistant but as a participant in high-end mathematical problem solving under expert human guidance. If it fails, the episode will still define the new verification culture around AI-generated mathematics: public code first, explanation later, and a field forced to decide how much proof is enough before understanding catches up.

Developments

  1. Anthropic's Levent Alpöge says Claude helped prove Riemann hypothesisQuartz · Nov 20, 2026, 8:22 PM · 7/10
  2. Anthropic's Levent Alpöge says Claude helped prove Riemann hypothesisQuartz · Nov 20, 2026, 8:22 PM · 9/10
  3. Anthropic Researcher Says Claude Helped Find Proof of Riemann Hypothesisbloomingbit · Nov 20, 2026, 5:51 PM · 7/10
  4. Anthropic Researcher Says Claude Helped Find Proof of Riemann Hypothesisbloomingbit · Nov 20, 2026, 5:51 PM · 7/10
  5. Who is Levent Alpöge? The Anthropic mathematician who says he and Claude proved the Riemann hypothesisNacionale · Nov 20, 2026, 5:47 PM · 7/10
  6. Who is Levent Alpöge? The Anthropic mathematician who says he and Claude proved the Riemann hypothesisNacionale · Nov 20, 2026, 5:47 PM · 7/10

Sources from the last 72 hours

  1. [1]Using AI, mathematician claims to have solved one of math’s great mysteriesNov 20, 2026, 11:45 PM
  2. [2]Transformation | Quanta MagazineNov 20, 2026, 1:00 AM
  3. [3]Unforced Euler singularity, unforced Navier-Stokes singularity, and the Generalized Riemann HypothesisNov 20, 2026, 1:00 AM
  4. [4]Hurried Announcement: Alessandro Masullo GRHNov 20, 2026, 1:00 AM
  5. [5]A Claimed Proof of the Riemann HypothesisNov 20, 2026, 1:00 AM
  6. [6]Anthropic Researcher Claims to Solve Riemann Problem With ClaudeNov 21, 2026, 1:17 AM
  7. [7]Anthropic researcher says AI chatbot Claude solved math’s greatest riddleNov 20, 2026, 9:00 PM

AI-generated article based on recent web research, then preserved as a dated editorial snapshot.