Overview
- Anthropic announced on Monday that an unreleased research version of its Claude model increased the proven lower bound for nontrivial Riemann zeta zeros on the critical line from 41.6% to 67.2%.
- The company says two Anthropic mathematicians examined Claude’s paper, external experts Brian Conrey and Dan Goldston gave rapid inspection, and the result has a machine‑checkable Lean formalization reported by Anthropic.
- Anthropic describes the work as produced in Claude Code over two sessions using about 31 million output tokens, with roughly 60 coordinated subagents running thousands of checks, scripts, and numerical tests.
- The finding is not a proof of the full Riemann Hypothesis and therefore carries no Millennium Prize claim; Anthropic attributes the advance to combining recent human research (Baluyot, Goldston, Suriajaya, Turnage‑Butterbaugh) with Bombieri’s earlier work.
- Experts and commentators urge independent verification and public disclosure of the technical proof and computations before the mathematics community accepts the claim, noting that reproducibility and fuller peer review remain the next necessary steps.