Loading today’s stories
Leadership2 distinct publishers3 min readPublished
The build fails unless the theorem rests on exactly Lean's three standard axioms, and an independent kernel written in Rust re-checked more than a million declarations. The eleven-day timeline rests on Anthropic's own account.
The Board Room · Leadership desk
Compiled by The Board RoomSomething wrong?How this is made
The check that does the most work is the comparison. leanprover's comparator v4.33.0 checked the build against a challenge file that states Fermat's Last Theorem using only Mathlib, and confirmed that the proved statement and every constant in it are identical to the challenge, and that the whole proof, Mathlib included, replays through the Lean kernel; the verdict recorded is "Your solution is okay!" [5]. That prerequisite is the part most work cannot borrow: someone had to write down the goal independently of the proof, in a form the artifact could be diffed against.
Independence also has a cost, which the repository discloses. The Rust kernel was built with four of the authors' own patches, one adding progress output and three speeding up its definitional-equality search, without which a few declarations occupy unmodified nanoda for many hours each; the repository states that none of them adds, removes or weakens a typing rule [7]. That disclosure does not undercut the purpose of running a second kernel. Lean 4.33.1, the toolchain used, itself carries the 2026 kernel soundness fixes [23], and a checker sharing no code with Lean's is how a kernel bug stops being invisible.
Volume explains why the audit had to be mechanical. Anthropic says Claude wrote 13 million lines of Lean and proved 29,500 intermediate theorems [15], which over eleven days is roughly 1.2 million lines a day [20]. Wiles's proof, published in May 1995, ran to 129 pages and needed months of painstaking work to verify [16]; the Lean artifact is about 100,000 lines for every one of those pages [21]. The repository does not pretend a reader closes that gap: its 390 MB html folder gives a page for each of 29,511 theorems and 1,450 definition modules, with English summaries generated automatically and the Lean statement authoritative [11].
The axiom guard shows only that a theorem carrying a particular name depends on three axioms; it says nothing on its own about whether those 29,511 intermediate results mean what they are called. The repository says the same: the checks establish that the statement follows from the three axioms given trust in the kernel and the checking tools, no tool can check that each intermediate theorem means what its name suggests, and that judgement is left to the reader, with PROOF-PATH.md naming the Lean theorem behind each step [9]. The narrowness is what makes the top claim carry weight. The statement uses Lean's built-in natural numbers, its one Mathlib ingredient is exponentiation on the naturals, which Mathlib defines as Lean's own, and nothing else in Mathlib has to be trusted because the kernel checks everything beneath [10].
The board-deck version is that a model proved Fermat's Last Theorem in eleven days. What an outside party can re-run is the proof; the eleven days and the working largely autonomously are Anthropic's account of its own process [14], and the two sit on different footing in the same release. The repository is a research artifact, not maintained and not accepting contributions [12], so it is a completed audit rather than a standing pipeline. For a buyer this quarter, the discrimination worth carrying into vendor conversations is between claims a build can fail on and claims that appear only in a document, and this release contains one of each.
Ranked by verification strength, evidence, and original report placement.
Anthropic says that along the way Claude wrote 13 million lines of Lean and proved 29,500 intermediate theorems.
The anthropics/fermats-last-theorem repository contains a complete, machine-checked proof of Fermat's Last Theorem in Lean 4 built on Mathlib (Lean 4.33.1; Mathlib v4.33.0, pinned by commit in lakefile.lean), following the argument of Frey, Serre, Ribet, Wiles and Taylor-Wiles.
The default build target FinalCheck.lean runs #print axioms fermat_last_theorem inside #guard_msgs expecting the list [propext, Classical.choice, Quot.sound], so the build fails unless the proof rests on exactly Lean's three standard axioms (no sorry, no added axiom, no native_decide).
FinalCheck.lean also derives Mathlib's own statement, FermatLastTheorem, from the repository's fermat_last_theorem.
A from-scratch lake build on Lean 4.33.1 with Mathlib compiled from source built all 60,475 modules of the repository, and every declaration was checked by the Lean kernel.
leanprover/comparator v4.33.0 checked the build against verification/comparator/Challenge.lean, which states the theorem using only Mathlib, and confirmed that the proved statement and every constant it mentions are identical to the challenge, that no other axiom is used, and that the whole proof, Mathlib included, replays through the Lean kernel; verdict: "Your solution is okay!"
Distinct publishers with included, body-backed reporting in this cluster.
1 article · September 4, 2026
1 article · September 4, 2026
Follow any of these and your For You feed starts watching them — no settings page required.
product
Anthropic formalized Wiles' proof in 11 days by handing the review to a machine1 distinct publisher
build
Palomar registers Lean proofs against a commit, and refuses to referee them1 distinct publisher
science
Text watermarks land on 2 December. The detection they imply does not.1 distinct publisher
build
Perf work stopped being a specialist queue item, and slow endpoints became a choice1 distinct publisher
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
Correctness reruns; the process does not
The mathematical claim is arranged so that a stranger can break it: the axiom list is asserted inside the default target, comparator replayed the entire proof including Mathlib against a challenge statement written only in Mathlib terms, and a second kernel implemented in Rust accepted 1,052,234 exported declarations. Two caveats sit in the repository's own text rather than in any critic's: nanoda ran with four patches written by the proof's authors, and no tool can confirm that a theorem named for a classical result proves that result. The eleven days and the agent count have no equivalent trail.
Published, awaiting outside builders
One named mathematician has read it. Kevin Buzzard says autoformalization artefacts are now robust enough to be built upon, but the repository he read declares itself unmaintained and closed to contributions, and our coverage shows no downstream user, no route into Mathlib, and no outside team that has repeated the five-and-a-half-hour build or the fifteen-hour replay.
Verification holds; autonomy is oversold
"First complete computer-checked proof" survives contact with the artifact better than most launch language does; the guard in the build target is a stronger commitment than a press claim. The stretch is in the word autonomously, which sits a few paragraphs from Anthropic's own record of Tianyi Peng telling Claude which theorem to push next, and in an eleven-day figure that the endorsing mathematician passes along as something Anthropic researchers say.
Author, checker-runner and publisher are one party
Anthropic wrote the proof, published the artifact, ran both checkers on its own hardware, and hosts the announcement that frames the result. The checking code is other people's, comparator from the Lean developers and nanoda from a separate author, which is the strongest structural counterweight in the story, though nanoda ran with the authors' patches. Until someone else spends the fifteen hours, everything here is one party's account of its own verification.
The theorem earns confidence; the timeline does not
The split is unusually clean. That the statement follows from Lean's three axioms is asserted in a form that fails loudly if false, on a toolchain carrying the 2026 kernel soundness fixes, and the repository is specific enough about memory, disk and wall clock that a determined reader could settle it. That Claude did it largely on its own in eleven days is a single company's account, and our number is dragged down by it.