Claude Raises Riemann Zeta Zeros to 67.2%: Two Papers No One Had Combined

Anthropic says an unreleased Claude research model raised the proven lower bound for the share of Riemann zeta function zeros on the “critical line” from 41.6% to 67.2%, a 25.6-point jump reached in about 36 hours. The company reported the model ran autonomously by coordinating roughly 60 AI subagents, generating 650 failed ideas before producing a cross-domain insight. Importantly, the advance did not resolve the Riemann Hypothesis itself; it still falls short of proving whether all nontrivial zeros lie on the critical line. However, Anthropic claims a verifiable, mechanically confirmed public Lean 4 proof is available on GitHub, while the model details are not disclosed. The prior best 41.6% bound held for decades before this result.