Anthropic AI tackles the 150-year-old Riemann hypothesis

The breakthrough occurred when an Anthropic staffer, lacking formal mathematical training, tasked the model with addressing the hypothesis. Over 36 hours, the system orchestrated 60 autonomous sub-agents to test 650 distinct approaches. Two agents surfaced the core mathematical insights, while 13 others validated the logic, eventually formalizing the result through the Lean proof assistant.

This event follows a surge in AI-led mathematical progress this year, including solutions to Erdos problems and the disproval of the Jacobian conjecture. While the results are mathematically sound—confirmed by Anthropic’s internal experts—they have triggered a broader philosophical rift. Some mathematicians fear the loss of accountability in discovery, as traditional proofs rely on specific authors taking responsibility for their work. Conversely, Fields Medalist Timothy Gowers argues that decoupling theorems from individual human creators may be a necessary, albeit complex, evolution for the field.