AI theorem provers upend math research pace

This is less a “math is over” moment and more a tooling inflection point that changes who can contribute and how quickly ideas get tested.

• –Research-level benchmarks like First Proof show AI is now being evaluated on unpublished, real mathematician-grade problems, not just textbook sets.
• –Formal systems such as Lean are becoming central because they let humans audit AI-generated proofs instead of trusting raw model output.
• –The bottleneck is shifting from proof generation to proof quality, interpretation, and integration into broader mathematical programs.
• –The near-term winners are likely hybrid human-AI teams, not fully autonomous math agents.