Researchers propose Scalable Delphi, a method for using large language models to estimate structured risk by adapting the Delphi method for LLMs with diverse expert personas, iterative refinement, and rationale sharing.