A company is evaluating the risk profile of deploying an AI agent with tool-use capabilities (file system access, email sending, code execution). Which principle from agentic AI safety frameworks MOST reduces the blast radius of a compromised or hallucinating agent?
Select an answer to reveal the explanation.
Short Explanation
Here's the deal — you should grant only the minimum permissions needed for the current task, then revoke them when it's done. That limits what a compromised or hallucinating agent can touch, so the blast radius stays small. Giving it all permissions upfront just makes the damage bigger.
Full Explanation
Least privilege is correct because it limits what a compromised or hallucinating agent can do by restricting tool permissions to only what the current task requires and making them revocable after completion, reducing maximum potential harm. Granting all permissions upfront maximizes blast radius. Publishing the permission list provides visibility but not prevention. Cross-checking multiple instances adds redundancy but does not constrain individual agent permissions.