Design Responsible AI, Security, Governance, Risk Management, and Compliance AB-100 Practice Questions (Page 1)

Questions 1–5

1. 1expert · hard

   A regulatory body requires a company to prove that its AI agent's decisions were made in accordance with its stated policies. The company has an audit trail of model changes but not of the agent's individual decisions. What should you implement to meet the requirement?

   • AUse a human reviewer to manually document each decision after the fact.
   • BStore only the final decisions without any context.
   • CLog every decision the agent makes, including the input, output, and the model version used.
   • DCreate a policy document that describes how decisions should be made.
   Reveal Solution Discussion Select an answer first
2. 2application · medium

   A nonprofit organization is deploying an AI agent to screen job applications. The organization is committed to diversity and inclusion. The responsible AI officer wants to ensure the agent does not discriminate against applicants based on gender, race, or age. What should the architect do to support this goal?

   • AConduct a fairness assessment using Azure Machine Learning's tools to evaluate the model's predictions across demographic groups, and document the results.
   • BRemove all demographic data from the application form so the model cannot see it.
   • CRely on the model's high accuracy as proof that it is fair.
   • DUse a black-box model and do not analyze its decisions to avoid bias.
   Reveal Solution Discussion Select an answer first
4. 3application · medium

   A company deploys an AI agent that answers employee HR questions using a vector index of internal policy documents. The security team is concerned about prompt injection via the agent's public chat interface. Which combination of controls should the architect implement to reduce the risk of the agent executing unintended actions or leaking sensitive data?

   • ADisable the agent's chat interface and require all queries to be submitted via email to reduce the attack surface.
   • BStore all grounding documents in a separate Azure subscription and grant the agent a shared access signature (SAS) token with write permissions.
   • CAdd input filtering to detect and block prompt-injection patterns, restrict the agent's permissions to read-only access on the grounding data, and enable audit logging of all agent interactions.
   • DIncrease the model's temperature to 0.9 so it is less likely to follow malicious instructions embedded in user prompts.
   Reveal Solution Discussion Select an answer first
5. 4foundation · easy

   Which practice is a common model security control?

   • AUsing a larger training dataset
   • BVersioning and hashing model artifacts to detect unauthorized changes
   • CDeploying the model on a public endpoint without authentication
   • DIncreasing the model's temperature setting
   Reveal Solution Discussion Select an answer first
6. 5application · medium

   A global logistics company uses an AI agent to process shipment documents. The company has customers in the EU and Asia. EU customer data must not leave the EU. The agent is deployed in a single Azure region in the EU. What should you do to ensure EU data residency while still serving Asian customers?

   • ARoute EU customer data to the EU region and Asian customer data to a separate region in Asia.
   • BStore all data in the EU region and serve Asian customers from the same region.
   • CEncrypt all data and store it in the EU region.
   • DUse a VPN to connect Asian customers to the EU region.
   Reveal Solution Discussion Select an answer first