Anthropic Claude Finds Elliptic Curve Rank 30, Breaking Decade-Old Math Record

Anthropic researchers used Claude to discover an elliptic curve of rank 30, breaking a ten-year record previously held at rank 29. The result was achieved through AI-assisted exploration of candidate curves, not through brute-force computation alone.

This validates LLM agents as viable research instruments for problems where search spaces are too vast for traditional heuristics but where structured, verifiable feedback exists. For AI operators, it signals that agentic loops can be productively applied to rigorous, formal domains — not just code generation or retrieval. The second-order effect is a workflow shift: hypothesis generation and candidate filtering become the LLM’s job, while human mathematicians focus on proof validation and interpretation. This makes high-level number theory exploration cheaper for institutions with model access, and it suggests that any field with formal verification (cryptography, lattice problems, combinatorial optimization) is now a candidate for agent-driven discovery. Builders should prioritize integrating formal checkers into agent loops, as that is the critical constraint enabling this class of output.