Lean and Mathlib certify the Claude-assisted Riemann proof — then a new challenge questions the geometry
On Saturday, November 21, the Lean and Mathlib communities jointly certified the Claude-assisted proof of the Riemann Hypothesis as correct, and MIT's Andrew Sutherland confirmed 175,187 modules compile with zero errors — but a single unverified post on November 22 claims the proof's geometry is trivially defined, and the Clay Institute still lists the problem as unsolved.

The News
On Saturday, November 21, the Lean and Mathlib communities issued a joint statement, endorsed by the Lean FRO and the Mathlib Initiative on the Claude-assisted proof of the Riemann Hypothesis: "We hereby certify that both the statement and the proof of the Riemann Hypothesis are correct." The proof was first claimed on Friday, November 20, by an Anthropic employee working with Claude. This post files only what is new since then: the joint certification on Saturday and the updated verification record on Sunday. Anthropic itself has made no announcement, so ANT is used here only as a read-through for the company behind Claude.
The verification record comes from MIT mathematician Andrew Sutherland, whose verification page was last updated at 04:35 UTC on Sunday, November 22. It confirms the published Lean source compiles without errors across 175,187 modules, the five headline theorems depend only on Lean's three standard axioms, and every module has been replayed through the Lean kernel with zero rejections. The independent checkers nanoda and con-leche accept the full exported proof, lean4lean passes every compiled module, and the compare modes of con-leche and its Rust port con-ron each confirm all five theorems. An independent sandboxed lake build reproduces the export byte for byte.
A semantic cross-check that does not rely on Mathlib's own definition of the zeta function also passed: combining the proof with the independent Robin1984 formalisation yields the Robin and Lagarias inequalities, with the comparator on the combined export passing at 04:16:52 UTC on November 22. Two checks are still in flight: the end-to-end lake comparator run and the Lean-kernel replay of the export. Separately, Stanford professor Ravi Vakil posted a draft proof of the generalised Riemann Hypothesis by Stanford postdoctoral researcher Alessandro Masullo the same day, per phys.org — a 940-page manuscript twelve years in the making, vetted for months by Brian Conrad. Within about eight hours, problems were found in that draft, and Vakil said it would not be accepted until the community was satisfied they were resolved.
The new counter-claim arrived on Sunday, November 22, in a single DEV post by Frank Chu, citing a paper by Jenny Lorraine Nielsen. Chu claims the generalised Riemann hypothesis portion of the proof is vacuous because the geometry predicate the proof uses is defined as "fun _ _ _ => True" in 24 files, while the real geometry predicate sits unused in 12 files. Chu says he opened the source and the definition is where the paper says it is — but with no second outlet or independent confirmation, this is filed here as an unverified counter-claim, not a fact. It targets the generalised portion's geometry; it does not establish an error in the Riemann-hypothesis theorem itself. The Clay Mathematics Institute still lists the problem as unsolved: its rules require publication, two years and general acceptance, and no human-readable paper exists yet.
Certified by machines, unjudged by people — and now challenged by a single post.
Why It Matters
This matters because a machine-checked proof is a different kind of evidence from a human-readable one. Lean is a proof assistant: a program that checks every logical step of a mathematical argument, so a proof that compiles with zero errors has survived a check no human referee can match. The joint certification, plus Sutherland's independent verification page, is the strongest signal yet that the formal statement has been proved — as computer-checked mathematics.
But certification is not the same as acceptance. The Clay Mathematics Institute, which offers a $1 million prize for the Riemann Hypothesis, still lists the problem as unsolved, and its rules require the proof to be published, to wait two years, and to win general acceptance from mathematicians. No human-readable paper exists yet, and most mathematicians cannot yet read the argument in the form it exists in — roughly 175,000 modules of Lean code. Until a readable version appears and survives scrutiny, the result lives in a middle ground: certified by machines, unjudged by people.
The November 22 challenge shows why that middle ground is fragile. If the geometry predicate really is defined as trivially true, the generalised portion could be vacuous — proving nothing about the geometry it claims to describe. But the claim rests on a single post by a single author, and no second source has confirmed it. The honest reading is that the challenge is a serious-looking question that has not yet been verified, and the certification, while verified, does not yet amount to a solved problem in the eyes of the Clay Institute.
Analyst's Take
Our call is Mixed. The joint certification and the verification record are real and significant — 175,187 modules, zero errors, independent checkers agreeing — but the Clay Institute still lists the problem unsolved, no human-readable paper exists, and a fresh, unverified challenge targets the generalised portion. Certification without acceptance is progress, not a verdict.
What to watch next is the two checks still in flight — the end-to-end comparator run and the kernel replay of the export — and whether anyone independently confirms or rebuts the Nielsen challenge. A second source either way would move this story more than more machine checks. Also watch for a human-readable paper, the missing piece for the Clay prize process and for mathematicians who want to judge the argument themselves.
For investors, the read-through is narrow. Anthropic made no announcement, and the proof's value to the company is reputational: a demonstration that its models can do frontier mathematics, if the result holds. ANT fell 8.50% to $88.91 on Friday in the broader market rout, and nothing in this weekend's certification changes the company's revenue. Reader poll: Will the Riemann proof win general acceptance — yes or no? Vote: Yes, it will hold, No, it will fail, Too early to tell.
Are you bullish or bearish on this story?
By The Numbers
FAQ
What is the Riemann Hypothesis?
It is one of the most famous unsolved problems in mathematics, about where the zeros of a function called the Riemann zeta function lie. It has resisted proof for more than 160 years, and the Clay Mathematics Institute offers a $1 million prize for solving it.
What is Lean?
Lean is a computer program called a proof assistant. Mathematicians write proofs in its language, and Lean checks every logical step automatically. If a proof compiles with zero errors, every step has been verified by the machine.
What is the Clay Mathematics Institute?
It is the organisation behind the Millennium Prize Problems, seven famously hard maths problems each carrying a $1 million prize. Its rules require a proof to be published, to wait two years, and to win general acceptance before the prize is awarded.
What does 'read-through' mean?
A read-through is a listed company used as a rough stand-in for a private company in a story. Anthropic is private, so its stock does not exist; ANT is the closest listed proxy, and its price moves for many reasons besides this proof.
👁
Source: FourWeekMBA