Effective permissions are everything an identity can actually do once all its roles, inherited rights and connected tools are added together. An account may have access from several sources, such as direct grants, group membership, inherited rights and the tools it can call, and its real power is the sum of them.

It matters because reviewing one policy at a time can miss the whole picture. A permission that looks harmless in one place can combine with another to allow something nobody intended, and an AI agent's reach includes what each of its tools can do on its behalf.

Working out the full set is not always easy. Systems differ in how they combine rules, some denials override grants and others do not, and tools provided by cloud platforms can simulate or report effective access, though their results depend on what they can see. The figure also changes as roles and tools change.

To use the idea, check what an identity can do in practice, use the platform's analysis tools where they exist, test the result by trying actions that should be refused and review it regularly.