Two mathematicians have independently claimed proofs of the Riemann Hypothesis, one of the most famous unsolved problems in mathematics, within a few hours of each other. Levent Alpöge, a mathematician at Anthropic, says he found his proof while working with Claude, the company's AI model. He released it not as a conventional paper but as Lean code, a formal proof language, totalling about 1.3 GB, which the article estimates would run to roughly 250,000 pages in prose. The work reportedly began on October 6, and the final step, labelled GRH, was posted on November 16. Alpöge says the method also covers the Generalized Riemann Hypothesis. He credits Claude with the difficult sections on automorphic forms on higher-rank groups, an area where he says he has little expertise. The Claude model used and the degree of human direction have not been disclosed.
Verification is the central open question. Elliot Glazer has confirmed that the formalized statement is GRH and puts the chance of a bug in the Lean code at under 1%, though the code was still compiling when the article was published. If it compiles without extra axioms, it is probably correct, but no person is likely to read all of it. In a separate announcement, Alessandro Masullo released a roughly 940-page proof of the hypothesis written entirely by humans. It is a work in progress that he says took him 12 years, and it had been undergoing private manual review. Commentators have contrasted the two: the AI proof is probably correct but close to incomprehensible, while the human proof is readable but probably contains technical errors. Both AI-claimed Millennium Prize results remain unverified and will take years to check. OpenAI's September claim on the Navier-Stokes problem is cited as a cautionary case, since critics argued its Lean proof did not match the natural-language result. ANT stock reportedly fell about 10% on the day, though the article does not link the move to the proof.
For Australian businesses and professionals, the proof changes nothing in day-to-day operations, costs or compliance, and it has no direct effect under the Privacy Act. The more practical lesson concerns verification. When AI produces work that people cannot realistically read, the reliable approach is to depend on checks that can be run mechanically, such as formal proof checkers or automated test suites, rather than on human reading. Teams that use AI for analysis or code should apply the same discipline to their own outputs.
Key Points
Anthropic mathematician Levent Alpöge claims a Claude-assisted proof of the Riemann Hypothesis and the Generalized Riemann Hypothesis.
The proof was released as roughly 1.3 GB of Lean code rather than a conventional paper, so it is very hard for humans to read.
Elliot Glazer confirms the formalized statement is GRH and puts the chance of a Lean bug under 1%, pending full compilation.
Alessandro Masullo has separately released a roughly 940-page, human-written proof that is still a work in progress.
Both AI-claimed Millennium Prize results remain unverified, and checking them may take years.