Should You Trust an Agent?

End-of-lesson check

15 questions · take it digitally for instant feedback at tendril.neural-forge.io/learn/quiz/end-builders-agentic-agent-trust

1. An AI agent that can only read files on your computer would be classified as what level of risk?

   1. No risk at all
   2. Low risk
   3. Medium risk
   4. High risk

2. Which of these is NOT one of the three trust principles mentioned in the content?

   1. Verify before trusting
   2. Give agents unlimited access initially
   3. Be ready to revoke trust quickly
   4. Match the trust level to the action

3. What does the principle of 'least privilege' mean when applied to AI agents?

   1. Least privilege means agents should never be trusted
   2. Agents should only have the minimum access needed for their specific task
   3. Agents need administrative rights to function properly
   4. Agents should always be given maximum capabilities

4. What is a 'sandbox' in the context of testing AI agents?

   1. A secure container where the agent can be tested without affecting real systems
   2. A physical area where robots are tested
   3. A type of computer game for training AI
   4. A sandbox is another name for the agent's memory

5. What does it mean to 'revoke' trust in an AI agent?

   1. To increase the agent's permissions temporarily
   2. To take away or remove the agent's access or capabilities
   3. To test the agent again after a failure
   4. To share the agent with other users

6. According to the trust framework, what should you do before giving an agent a new, more powerful capability?

   1. Give it full access immediately to see what it can do
   2. Ask the agent if it thinks it can handle it
   3. Test it thoroughly in safe situations first
   4. Only give new capabilities to paid versions of agents

7. An agent that can 'draft' content but cannot 'send' it is what level of risk?

   1. No risk because drafting is not dangerous
   2. Medium risk because it could create harmful content
   3. Low risk because a human must approve before sending
   4. High risk because it can create content

8. What is the recommended approach for expanding what you trust an agent to do?

   1. Ask other people what permissions to give
   2. Only use agents for one specific task forever
   3. Give it full access on day one
   4. Start with low-risk tasks and expand carefully

9. Which of these pairs correctly shows lower risk to higher risk?

   1. Sending posts → Drafting messages
   2. Reading emails → Writing checks
   3. Searching the web → Reading files
   4. Making decisions about people → Summarizing documents

10. A student uses an AI agent to help write a school essay. The agent suggests sentences but the student picks which ones to use and types them in personally. What risk level is this?

    1. Low risk because the student makes all final decisions
    2. No risk because it's just schoolwork
    3. Medium risk because the AI influences the content
    4. High risk because the AI is doing schoolwork

11. If an AI agent has been making good decisions for months, what should you do before trusting it with something new and riskier?

    1. Test it in a safe way first to verify it handles the new capability well
    2. Give it the new task immediately to save time
    3. Refuse to give it any new capabilities ever
    4. Trust it completely since it has proven reliable

12. What makes 'spending money' a high-risk activity for AI agents?

    1. Money spending is not actually high risk
    2. It can result in direct financial loss and is hard to undo
    3. AI cannot count money accurately
    4. It requires a credit card number

13. A company wants to use an AI agent to screen job applicants. According to the framework, what should they consider?

    1. This is low risk since it's just looking at resumes
    2. This is the highest risk category because it makes decisions about people
    3. This is medium risk and needs careful monitoring
    4. This should never be done by AI

14. Why is 'matching trust level to action' important?

    1. It prevents giving too much power for simple tasks, limiting potential harm
    2. It makes the agent more intelligent
    3. It is required by law
    4. It makes agents work faster

15. What should you do immediately if an AI agent starts behaving unexpectedly or exceeding its boundaries?

    1. Wait and see if it improves on its own
    2. Give it more time to learn
    3. Revoke its trust immediately
    4. Ask it to explain why it misbehaved