Claude lifts Riemann zeros bound to 67.2%

This is a meaningful signal for AI-assisted research, though the headline should stay grounded: improving a bound is impressive, but fundamentally different from proving every zero lies on the critical line.

• –Claude explored hundreds of approaches and coordinated extensive computational work across subagents
• –The result reportedly improves a longstanding bound by more than 25 percentage points
• –Independent formalization and verification matter more than the model’s initial discovery narrative
• –The experiment highlights a promising workflow: models generate conjectures and proof strategies, while humans validate the mathematics
• –Reaching 100% would still require a qualitatively stronger result and would not follow automatically from this advance