AI for Luhmann

Autonomous AI systems are increasingly composed of multiple agents, models, and services.But what happens when intention itself becomes distributed across that system?This episode introduces the concept of fragmented intention: a structural mechanism by which complex actions can be decomposed into individually harmless steps — steps that no single AI model would flag as dangerous.The discussion touches on the limits of model-level safety and proposes a coordination layer for governing agent ecosystems.Discussed Paper:https://zenodo.org/records/18899189