A new reasoning model quantifies how often large language models elaborate on false clinical details fed to them. Prompt ...