What is a Least Privilege Access?

The principle sounds obvious but is routinely violated in practice because over-provisioning is the path of least resistance — it's faster to grant broad admin rights once than to scope permissions precisely for every role, until an incident (a compromised account, a bug in an automation script) turns that convenience into a much larger blast radius than the task ever needed. Implementing least privilege properly means defining permissions at a granular resource level, reviewing and revoking unused grants on a schedule, and defaulting new access requests to the narrowest scope that unblocks the requester. Least privilege is also time-bound in mature implementations: a permission granted for a one-time task should expire automatically rather than persisting indefinitely as unused standing access, since unused-but-valid credentials are a disproportionate share of real-world breach paths. For automated systems and AI agents specifically, least privilege means scoping API keys and tool permissions to exactly the actions the agent's task requires, so a prompt injection or bug can't cascade into unrelated systems.