Observability for Delegated Execution in Agentic AI Systems
arXiv·medium signal
Identifies a sharp gap: delegation-scoped execution is not identifiable from standard observables — audit logs and execution traces can be identical under multiple incompatible delegation assignments. This makes it impossible to tell which agent was actually authorized to act. A foundational concern for anyone building auditable, compliant multi-agent systems where 'who did what on whose behalf' must be provable.