Which limit reduces harm if an agent is manipulated?
Why C is the answer
Least-privilege tool permissions
An agent can only misuse the access it holds.
In detail
- An agent can only misuse the access it holds.
- Assume manipulation is possible.
Use case
An illustrative scenario, not a real incident.
A manipulated agent has read-only access, so it cannot do much harm.
See it in different ways
flowchart LR A["Input checks"] --> B["Least-privilege tools"] B --> C["Output checks"] C --> D["Audit log"]
sequenceDiagram participant W as Retrieved page participant A as Agent participant P as Tool permissions W->>A: Hidden instruction A->>P: Attempts an action P--xA: Outside its permissions
flowchart TD
Q{"Does the agent need this access?"}
Q -->|"Yes"| A0["Grant it, narrowly"]
Q -->|"No"| A1["Do not grant it"]Key point: Least privilege limits harm.
Why the other options are wrong
- A. A longer retention for its context
- Retention does not limit access.
- B. A broader API key shared by all tools
- A broad key raises the risk.
- D. A higher token limit for each call
- Tokens do not limit access.