Topics

The vocabulary of governing AI agents

The security industry has spent twenty years building a vocabulary for humans and service accounts: identity, access, permissions, least privilege. AI agents reuse most of those words and quietly change what they mean. An agent with a valid credential and a valid permission can still make a decision nobody authorized it to make.

These pages define the terms one at a time. Each one starts with a plain definition, works through a concrete example, and ends with where the concept stops and what has to sit next to it. Where we are stating a position rather than a settled definition, we say so.