PERMISSIONS AND ACCESS CONTROL

CHECK YOUR UNDERSTANDING • Why should deny rules be checked before allow rules? ◦ Answer: So a specific prohibition cannot be overridden by a broader permission. • Why are allow-lists usually stronger than deny-lists? ◦ Answer: Because you cannot list every dangerous item in advance, and a deny-list leaves open anything you forgot. • Should human approval depend on how confident the agent sounds? ◦ Answer: No, it should depend on objective risk triggers, since a confident agent can still be wrong or manipulated. • Why keep API keys out of the model's context? ◦ Answer: Anything the model sees may be repeated, logged, or leaked through injection. • What should happen if a key is exposed? ◦ Answer: Revoke it, issue a new one, check logs for misuse, and notify the owners.