C1000-185 · Drill mode

In an agentic AI system that calls tools (APIs/functions), which mitigation reduces the risk of an LLM making unintended destructive calls?

• AA. Enforce least-privilege scopes on tool credentials, require human-in-the-loop confirmation for high-impact actions, and add allow/deny rules
• BB. Give the agent the root credentials
• CC. Run all tools without any guardrails
• DD. Remove logging