Agents need Authorization, not just Authentication

As AI agents become increasingly integrated into enterprise infrastructures, traditional access control models are being challenged, particularly by the "Confused Deputy" problem, where agents unintentionally misuse their extensive privileges. This has prompted a reevaluation of agent identity in the Identity and Access Management (IAM) stack, leading to innovations like Microsoft's Entra Agent ID and IETF's SCIM resource standardization. Fine-Grained Authorization (FGA) is emerging as a solution, providing a hierarchical extension of Role-Based Access Control (RBAC) that allows for precise, context-specific permissions. FGA enables dynamic authorization checks, preventing both "God Mode" access and inadvertent data leaks by ensuring that agents operate within strictly defined scopes. As agents evolve from passive scripts to active collaborators with memory, securing long-term context and managing lifecycle governance are critical, requiring centralized governance over agent identities and their associated permissions. Research is ongoing to address complex authorization challenges, such as multi-hop delegation and Intent-Based Access Control, aiming to align agent actions with organizational policies and initial user intents.