Reply to "When do large language models cross the line: "reasoning" red teaming in healthcare".

Abstract

We appreciate Sorin et al. for highlighting critical considerations for future red teaming of large language models (LLMs) in healthcare. We agree that analyzing only final answers overlooks failures in internal reasoning and that reasoning models introduce new risks. Expanding red teaming to assess reasoning quality, cognitive biases, and false consistency, as well as adopting ethically varied scenarios, will strengthen LLM auditing frameworks.