Design Responsible AI, Security, Governance, Risk Management, and Compliance AB-100 Practice Questions (Page 8)

Questions 36–40

1. 36application · medium

   A nonprofit organization deploys an AI agent that recommends job training programs to unemployed individuals. The organization wants to ensure the agent does not discriminate based on age, gender, or ethnicity. What should you do as part of a responsible AI review?

   • AAsk the agent to provide a disclaimer that its recommendations may be biased.
   • BTest the agent with diverse input data and evaluate the recommendations for bias across protected groups.
   • CLimit the agent to recommending programs that are popular among all users.
   • DRemove all demographic data from the training data to ensure fairness.
   Reveal Solution Discussion Select an answer first
2. 37foundation · easy

   What is the purpose of a responsible AI review?

   • ATo evaluate AI solutions against ethical principles and standards
   • BTo optimize the model's inference speed
   • CTo reduce the cost of model training
   • DTo measure the model's accuracy on a test set
   Reveal Solution Discussion Select an answer first
4. 38application · medium

   A government agency deploys an AI agent that assists citizens with benefit applications. The agency requires that the agent's decisions be explainable and that any changes to the agent's behavior be reviewed by an ethics board. What governance framework should you establish?

   • ADefine an approval workflow for any change to the agent's prompts, models, or grounding data, with the ethics board as an approver.
   • BUse a separate model for each citizen to avoid bias.
   • CCreate a policy that requires all agent responses to be reviewed by a human before sending.
   • DAllow the development team to deploy changes freely but log all changes for later review.
   Reveal Solution Discussion Select an answer first
5. 39foundation · easy

   Which of the following is a primary security threat that agent-specific design should mitigate?

   • AHigh token consumption increasing cost
   • BPrompt injection attacks that manipulate the agent's behavior
   • CNetwork latency causing slow responses
   • DLack of user-friendly interface
   Reveal Solution Discussion Select an answer first
6. 40expert · hard

   A tech company's AI agent is vulnerable to a prompt-injection attack that causes it to execute unauthorized actions, such as sending emails. The agent uses a set of tools with permissions. The security team wants to mitigate the risk without completely disabling the agent's functionality. Which mitigation is most effective?

   • ARemove the email tool from the agent's toolset.
   • BAdd a system prompt that instructs the agent to ignore any instruction to send emails.
   • CUse a different model that is less susceptible to prompt injection.
   • DRequire human approval for any action that sends emails or has external effects.
   Reveal Solution Discussion Select an answer first