What is least agency?
Least agency is the principle of giving an AI agent only the autonomy, tools and access it needs for its task, extending least privilege from permissions to actions.
- Least agency limits what an agent may decide, not only what it may access.
- It is the companion to least privilege, not a replacement for it.
- The practical form is approval on the actions that are hard to undo.
- Marking everything as requiring approval destroys the control, because people stop reading.
Published Last reviewed
In practice
Least privilege limits what an identity is allowed to access. Least agency also limits what an agent may decide to do with that access: which tools it can call, which actions need a human, and how far it can act without review. OWASP describes the underlying risk as excessive agency.
In practice, least agency starts with knowing an agent’s data reach. You cannot limit autonomy sensibly without knowing what the agent could reach.
Least agency is the principle of giving an AI agent only the autonomy, tools and access it needs for its task, extending least privilege from permissions to actions.
Privilege and agency, side by side.
Two questions about the same agent. Most programmes answer only the first.
| Least privilege | Least agency | |
|---|---|---|
| The question | What may it access? | What may it decide? |
| Expressed as | Roles, scopes, permissions. | Approval gates and action limits. |
| Fails when | Permissions grow and nobody prunes. | Everything needs approval, so nothing is read. |
| Needed because | Credentials get stolen. | Agents get steered. |
How it works. Step by step.
The parts that make it up, in an environment with AI agents.
Give the agent only the tools its task needs.
Require a human for destructive or irreversible actions.
Use a dedicated identity with the narrowest access that works.
Check what the agent could reach before deciding how much autonomy it gets.
Questions, answered.
Short answers about Least agency.
What is the difference between least agency and least privilege?
Least privilege limits what an identity may access. Least agency also limits what an agent may decide to do: which tools it can call, which actions need a human and how far it can act on its own.
What does OWASP say about agency?
The OWASP Top 10 for LLM Applications lists excessive agency as a risk: an LLM-based system given more functionality, permissions or autonomy than it needs.
Where do you start with least agency?
With the agent’s data reach. You cannot decide sensibly how much autonomy an agent gets without knowing what it could reach if it acted on its own.
Keep reading.
Pages that belong to the same picture.
See what your agents can reach. Before someone else does.
Book a demo and we walk through the access paths in an environment like yours.