Anthropic's agents advanced Riemann hypothesis — without a human mathematician
an unreleased anthropic model advanced the riemann hypothesis by testing 650 ideas via 60 autonomous sub-agents, signaling a shift toward ai-discovered scientific truth.
anthropic revealed that an unreleased model improved the lower bound of the riemann hypothesis by coordinating 60 sub-agents to generate and validate 650 distinct mathematical approaches. the system operated autonomously for 36 hours, spending $31 million in compute to produce a proof subsequently verified by human mathematicians using the lean proof assistant. this breakthrough highlights the potential of multi-agent architectures to tackle complex scientific workflows, reigniting concerns within the mathematical community regarding attribution and the role of human researchers in future discoveries.