HomeTech & AIAnthropic Mathematician Claims an AI-Assisted Proof of the Riemann Hypothesis. Here’s What...

Anthropic Mathematician Claims an AI-Assisted Proof of the Riemann Hypothesis. Here’s What Has and Hasn’t Been Checked

An Anthropic mathematician says he and an AI system have produced a proof of the Riemann Hypothesis, one of the most famous unsolved problems in mathematics. Mathematicians say it is too early to call it settled, but one key check has already come back positive.

Levent Alpöge, who works at Anthropic, published an essay on Nov. 20 saying he found a solution to the 167-year-old problem that compiles in Lean, a proof-checking program. According to TechCrunch, several participants in a Lean community forum, including MIT mathematician Andrew Sutherland, said they compiled it successfully, and the article was updated to say independent sources had checked the Lean proof.

What the Riemann Hypothesis is

First posed in 1859, the hypothesis is one of the Millennium Prize problems. TechCrunch describes it as a conjecture that there is a hidden structure underlying the appearance of prime numbers, the whole numbers divisible only by 1 and themselves. Many results in number theory are built on the assumption that it is true, which is why a proof would matter far beyond one equation.

Why this one is different from a normal claim

Big math claims usually arrive as a paper that experts read line by line. Here, the proof is described by Terence Tao, the UCLA mathematician, as an “(enormous) Lean-only proof” of the Generalized Riemann Hypothesis, a broader version, generated by an internal AI model. Lean is software that checks each logical step mechanically. If the code compiles, the steps follow from the stated assumptions, which is why the early compile reports carry weight.

But a machine check is not the same as human understanding. TechCrunch notes that Alpöge’s proof lacks a human-readable explanation, and that he has committed to working toward one. Tao says his own very initial look suggests the proof uses the full power of modern developments in the Langlands program and arithmetic geometry, and he expects that “the digestion process will take some time.”

A human-written proof is also in play

The same week, NYU mathematician Alessandro Masullo, a Stanford postdoc, circulated a proof of the Generalized Riemann Hypothesis written by people alone. Per TechCrunch, Stanford’s Ravi Vakil, who posted it, said “it is all human, with no AI,” and Masullo says it took 12 years of work. That proof is not yet in Lean, so TechCrunch says it is too early to tell whether it checks out.

Two more AI-assisted results

Tao’s blog post lists two related results from Alpöge and Tristan Buckmaster, both with heavy AI assistance and both formalized in Lean: a proof that the 3D unforced Euler equations can develop a singularity in finite time (written up with Matei Coiculescu as a human-readable paper), and a similar proof for the Navier-Stokes equations, which Tao says still needs work to become a professionally written proof. These equations describe fluid flow, and the Navier-Stokes question is also a Millennium Prize problem.

The catch

Nothing here has been through formal peer review, and a compiled Lean file only proves what it was told to prove. Experts still need to confirm that the statements encoded in Lean match the real conjecture, and that the argument can be understood and reproduced. Alpöge himself conceded in his essay that “the precise timing is horrible,” according to TechCrunch. Expect weeks or months of scrutiny before the math community weighs in.

Why it matters even if it holds

If the proof stands, it would be among the first major open problems in pure math settled with heavy AI involvement, and the Lean check is what separates this from a vague boast. If it fails, the failure will likely be a mismatch between what was encoded and what the conjecture actually says, not a hidden arithmetic slip. Either way, the community’s process for checking machine-made proofs is being tested in public.

What to watch

  • Whether the Lean statements are confirmed to match the standard Riemann Hypothesis.
  • A human-readable write-up from Alpöge.
  • Whether Masullo’s human proof is formalized or confirmed.
  • Any formal statement from the Clay Mathematics Institute, which administers the Millennium Prizes. None had been reported in the sources we read.

More on Contoh

Sources