arxivcs.AIcs.CL2026-07-11
Information-seeking failures of large language models in agentic clinical reasoning
Krischan Braitsch, Laura K. Schmalbrock, Theresa Weltermann, Andrew F. Berdel, Isabella Miller, Kai Tran, et al.
Large language models achieve high scores on medical knowledge assessments, yet clinical reasoning requires actively deciding what to investigate under uncertainty. We developed an agentic evaluation framework in hematologic oncology in which models must proactively request clini…