Part 7 · Agentic AI Security
ASI03: Identity and Privilege Abuse
Delegation chains and inherited credentials can give an agent more authority than the person it serves.
What it is
Identity and privilege abuse exploits dynamic trust and delegation between agents. Attackers manipulate delegation chains, role inheritance, control flows, or agent context to escalate access or bypass controls.
Agents often lack a distinct, governed identity, so traditional access systems struggle to tell who is really acting.
How the attack unfolds
This is the confused deputy problem: a trusted intermediary uses its authority for someone who lacks it.
Common patterns
- Inherited privileges passed down a delegation chain.
- Credentials cached in memory or conversation history.
- Confused deputy requests between trusted agents.
- Stale permissions that remain after the task changes.
- Spoofed agent identity in multi-agent systems.
Defenses
- Issue task-scoped credentials. Keep them short-lived and narrow.
- Give each agent its own identity. Avoid shared service accounts.
- Preserve the user's identity. Carry it through every delegation.
- Isolate sessions. Clear context and credentials between tasks.
- Authorize every step. Re-check permission at each privileged action.
- Audit delegation. Record who authorized what, and through which agent.
Common misconception
Common belief
"The agent is authenticated, so its request is authorized."
In practice
Authentication proves who the agent is. It does not prove that the person behind the request has permission.
Key takeaways
- Delegation can pass along more authority than intended.
- Credentials should be short-lived and scoped to a task.
- Authorization must consider the original user.