Anthropic Model Lifts Riemann Hypothesis Lower Bound After Testing 650 Ideas

3 articles · Updated · TechCrunch · Aug 11

Anthropic said an unreleased AI model significantly extended the verified lower bound for cases where the Riemann hypothesis holds, a notable advance on a 150-year-old problem carrying a $1 million prize for a general proof.
Over about 1.5 days, the system explored 650 approaches through 60 sub-agents and 31 million total compute steps after a non-specialist staffer prompted it to seriously attempt the proof.
Two sub-agents produced the key mathematical ideas, while others supplied suggestions, checked arguments and helped draft the paper; two in-house mathematicians confirmed the result and Lean formalized it.
The result adds to a run of AI-led math advances, including OpenAI's 10 major Astra proofs and an Anthropic effort that disproved the Jacobian conjecture.
That momentum is splitting mathematicians: a June declaration warned AI could weaken author accountability for proofs, while Fields Medalist Timothy Gowers argued the technology may reshape the field in productive ways.

Sources

Left100%

TechCrunch9d ago

Anthropic AI Model Advances Riemann Hypothesis, Increasing Solution Lower Bound

AI Insider9d ago

Anthropic Says Claude Improved a Longstanding Bound Tied to the Riemann Hypothesis

YouTube9d ago

AI makes progress on the Riemann Hypothesis! Huge improvement on lower bound of zeroes on the line!

11 Sources

If an AI solves a 150-year-old math problem after a casual prompt, who truly deserves the credit for the discovery?

Could the reliance on machine-verified proofs mean humans will soon stop understanding the very mathematics they are publishing?

How did 60 AI sub-agents avoid cascading hallucinations while autonomously orchestrating millions of tokens to advance frontier mathematics?

Claude AI Raises Riemann Zeta Critical Line Bound to 67.2%: A Historic Leap in AI-Driven Mathematics

Overview

In August 2026, Anthropic's AI model Claude achieved a historic mathematical breakthrough by autonomously raising the proven lower bound of Riemann zeta function zeros on the critical line from 41.6% to 67.2%. This leap was made possible after Claude, initially prompted by Jarred Sumner and encouraged to persist despite 650 failed attempts, deployed around 60 subagents to rigorously explore, test, and critique mathematical ideas. By analyzing complex quadratic forms without simplifying assumptions, Claude captured subtle interactions between zeros, leading to its record result. The proof was then formalized in Lean for absolute certainty, sparking excitement and cautious optimism in the mathematical community.

...