Agent Identity and the Permission Problem

An AI agent does more than execute predefined instructions. It interprets a goal and chooses actions, making identity and access controls central to its governance. This paper examines how named ownership, limited and expiring permissions, purpose constraints, logging, and revocation can control what an agent is allowed to touch and do.