Newzly.net

business

Anthropic Scientist Claims Riemann Hypothesis Proof Using 17 Million Lines of Claude AI Code

An Anthropic researcher has unveiled a 17-million-line formal proof of the Riemann Hypothesis, utilizing an unreleased version of Claude AI to tackle one of mathematics' greatest unsolved problems.

3 min readvia Divya Bhaskar Business
Visual representation of the Riemann zeta function alongside lines of mathematical code and artificial intelligence neural network patterns.
Visual representation of the Riemann zeta function alongside lines of mathematical code and artificial intelligence neural network patterns.

In this piece

An Anthropic mathematician has released a formal proof of the Riemann Hypothesis written in 17 million lines of Lean 4 code with assistance from Claude AI. If…

A researcher at artificial intelligence firm Anthropic has claimed to have developed a formal mathematical proof for the Riemann Hypothesis with the help of an unreleased version of the Claude AI model. The purported solution, announced by Anthropic mathematician Levent Alpöge, was released not as a conventional paper, but as an enormous codebase spanning approximately 17 million lines of machine-checked code. The announcement has ignited intense discussion across both the computer science and mathematical research communities.

The computational proof is written in Lean 4, an open-source programming language and interactive theorem prover designed to verify mathematical logic step by step. Spanning more than 175,000 individual files, the project represents an unprecedented effort to outsource complex analytical deductions to an automated verification engine. Alpöge estimated that translating the Lean code into a standard human-readable paper would require roughly 250,000 pages of dense technical text.

Formulated by German mathematician Bernhard Riemann in 1859, the hypothesis is widely regarded as one of the most profound open problems in pure mathematics. The conjecture concerns the distribution of prime numbers and posits that all non-trivial zeros of the Riemann zeta function have a real part equal to one-half. In 2000, the Clay Mathematics Institute designated it one of seven Millennium Prize Problems, carrying a one-million-dollar bounty for its first verified proof.

The breakthrough reportedly emerged from a hybrid collaboration where Alpöge provided high-level conceptual scaffolding before the frontier Claude model took over the bulk of theorem generation. As the work neared its conclusion, the AI began generating intricate structural proofs and lemmas that surpassed immediate human synthesis, according to project disclosures. The system completed technical arguments at a velocity and scale that human researchers could not match manually.

The milestone builds upon preliminary advances achieved earlier in the year when Anthropic applied Claude to establish sharper bounds on zeros along the critical line. Those earlier computational results hinted that large language models trained alongside formal verification tools could meaningfully assist in higher mathematics. However, expanding that capability to tackle the entire Riemann Hypothesis and its generalized counterpart marks a dramatic escalation in scope.

Early reactions from the academic community have centered on the monumental challenge of auditing the 17-million-line codebase. External mathematicians and computer scientists quickly set up automated environments to compile the Lean packages and test whether the exported kernel steps held up without logical contradictions. Initial reports indicate that the modular files successfully compiled and passed automated kernel checks, confirming that the code contains no syntactic or mechanical errors.

Despite the software compiling, mathematicians emphasize that mechanical verification is distinct from human conceptual understanding. A proof consisting of millions of lines of machine code creates an epistemological dilemma, as no individual scholar can read or comprehend the complete chain of reasoning. Experts note that this marks an inflection point where artificial intelligence may begin producing valid scientific knowledge that fundamentally outstrips human cognitive bandwidth.

Full institutional validation from bodies such as the Clay Mathematics Institute will require an extensive peer-review process that could take years. Traditional evaluation protocols demand that mathematical proofs be thoroughly scrutinized by domain experts to ensure that no hidden assumptions or circular definitions were baked into foundational axioms. For now, the global mathematical community is mobilizing independent toolchains to parse the code and verify whether the centuries-old riddle has truly been laid to rest.

Primary source: Divya Bhaskar Business. Newzly rewrites and structures source reporting for clarity; verify critical details at the original.