Scoped Access
Giving an agent the narrowest set of permissions that still lets it finish the job.
Think of it like
Handing a house-sitter the key to the front door and the alarm code — not the safe, the car, and your bank login.
Example
A summarizing agent gets read-only access to one folder, not write access to the whole drive, so a bug can’t delete anything.
How it actually works
Scoping is the principle of least privilege applied to agents: minimize the blast radius if something goes wrong or gets hijacked. It matters more for agents than for ordinary software because a prompt-injected agent will happily use whatever access it has. Good scoping is granular — per-tool, per-resource, per-task — rather than one big “can do everything” grant.
For product teams
Narrow access limits how much damage a confused or compromised agent can do — scope tightly by default.
For engineers
Grant per-resource, per-action tokens so a compromised agent’s reach is bounded to the task at hand.
Related
- Permissions — The broader system this narrows.
- Agent Guardrails — Fences that keep it inside its lane.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome