AI interpretability has long been the focus, but what if it’s only part of the story? New research introduces model semantics, a framework for understanding what AI systems truly represent and how their internal structures connect to real-world phenomena.