Definition
Plain language
Harm caused by the treatment itself rather than by the original problem.
As stated in the literature
A medical term used in this corpus to describe harm produced by an intervention's intended mechanism; Fukui applies it to alignment training that simultaneously increases hit rate and false-alarm rate via a single criterion shift.
Why it matters: The concept reframes some alignment side effects as caused by the training procedure itself, not just incidental, which changes how you fix them.
For example, a surgery that successfully removes a tumor but causes nerve damage in the process has produced an iatrogenic injury.
Heard on the show
“" Fukui reaches, carefully, for a medical word: iatrogenic.”Episode 087 — When No Agent Reads the Whole Document: A Universal Cliff in Multi-Agent Review