AI agent glossary
Least privilege
The security principle of giving an agent only the permissions and access it truly needs for its task, and nothing more.
Least privilege is an old idea from computer security that matters even more for agents. An agent can be confused, manipulated by prompt injection, or simply wrong. Whatever it can access is what it can damage.
In practice this means read-only credentials when the agent only needs to read, access scoped to a single folder or database table, API keys limited to specific actions, and tools that do one narrow thing rather than offering a general shell.
It also means removing access when it is no longer needed. A good test: if this agent went rogue right now, what is the worst it could do? Shrink that answer until you are comfortable with it.
A reporting agent gets a database user that can run SELECT queries on the analytics tables only, so even a malicious instruction cannot make it delete or modify data.
Learn more in the course
Related terms
Go from definitions to doing
Learn to use and build AI agents in 65 practical lessons.