Independent Verification Confirms AI-Generated Proof of Riemann Hypothesis Is Likely Valid

Mathematician Levent Alpöge from Anthropic used the Claude AI model to produce a proof for the Riemann hypothesis, one of mathematics' most significant unsolved problems. The proof, written in the Lean programming language, has been independently compiled and verified by researchers including Andrew Sutherland from MIT, with no errors detected. Experts now consider the proof's validity highly probable, though final confirmation of its alignment with the original theorem remains under scrutiny.
On November 20, 2026, mathematician Levent Alpöge of Anthropic announced that the Claude AI model had generated a proof for the Riemann hypothesis, one of mathematics’ most enduring unsolved problems since its formulation by Bernhard Riemann in 1859. The proof was encoded in the Lean programming language, a formal verification system designed to ensure mathematical correctness through computational checking. Independent verification by Andrew Sutherland of MIT confirmed that the Lean code compiled without error using only three standard axioms, a result that experts like Kevin Buzzard of Imperial College London described as “approximately zero” chance of being flawed. This milestone follows earlier skepticism about AI-generated proofs, particularly regarding whether the formal logic aligns with the original mathematical statement. The key unresolved issue is whether the proof genuinely addresses the Riemann hypothesis or a subtly different claim—a concern addressed through a “comparator run” that checks for hidden redefinitions or kernel-specific exploits. While the code has passed initial compilation, full validation requires independent runs across multiple Lean kernels to rule out undetected bugs. The broader mathematical community remains cautious, awaiting the final comparator run before accepting the proof as definitive.
If confirmed, this AI-generated proof could mark a turning point in mathematical discovery, blurring the line between human intuition and machine reasoning. The Riemann hypothesis underpins much of number theory and has implications for cryptography, computer science, and algorithmic complexity. Should the proof be validated, it may prompt a reevaluation of how mathematical knowledge is produced and verified, potentially accelerating research in formalized mathematics. Institutions like MIT and organizations such as Anthropic may gain increased influence in shaping the future of AI-assisted science. A comparable precedent is the 1976 proof of the four-color theorem, which relied on extensive computer calculations and initially faced skepticism before gaining acceptance. This case may similarly challenge the mathematical community’s traditional reliance on human-readable proofs. The next critical step—successful cross-kernel verification—will determine whether AI-generated proofs can be trusted as foundational. If so, formal verification systems like Lean may become standard in high-stakes mathematical research, possibly reshaping academic publishing, peer review, and even educational curricula. The broader impact could extend to how society views the role of AI in advancing human knowledge.