Design Responsible AI, Security, Governance, Risk Management, and Compliance AB-100 Practice Questions (Page 5)

Questions 21–25

1. 21foundation · easy

   Which mechanism is commonly used to control access to model tuning resources?

   • AUsing a larger batch size during training
   • BIncreasing the model's learning rate
   • CEnabling more verbose logging
   • DRole-based access control (RBAC) with least-privilege permissions
   Reveal Solution Discussion Select an answer first
2. 22expert · hard

   A multinational corporation wants to deploy an AI agent for employee HR services. The company has employees in the EU, the US, and Japan. EU data must stay in the EU, and Japanese data must stay in Japan. The company wants a single agent deployment to reduce cost. What is the best approach?

   • AStore all data in Japan and use a single agent there.
   • BDeploy a single agent in the EU and route all data to the EU.
   • CUse a single agent in the US and rely on encryption.
   • DDeploy separate agent instances in the EU and Japan, and route data based on the employee's region.
   Reveal Solution Discussion Select an answer first
4. 23foundation · easy

   What is the primary purpose of an audit trail for AI models?

   • ATo increase the model's token limit
   • BTo track and document changes to models and data for accountability
   • CTo reduce inference latency
   • DTo improve model accuracy
   Reveal Solution Discussion Select an answer first
5. 24application · medium

   A university is deploying an AI agent to assist students with course registration. The agent uses a knowledge base of course catalogs and student records. The university's IT department requires that only authorized students and staff can access the agent, and that all interactions be logged for security audits. What should the architect implement?

   • ADisable logging to protect student privacy, as logs may contain personal information.
   • BUse a single service account for all users and rely on the agent to identify users based on their queries.
   • CIntegrate the agent with Azure Active Directory (Azure AD) for authentication, use RBAC to control access to the knowledge base, and enable diagnostic logging for the agent.
   • DAllow anonymous access to the agent but restrict access to the knowledge base using a shared secret key.
   Reveal Solution Discussion Select an answer first
6. 25foundation · easy

   Which mitigation is most effective against prompt injection attacks?

   • AUsing a larger model with more parameters
   • BIncreasing the temperature to make outputs more random
   • CDisabling logging to reduce attack surface
   • DValidating and sanitizing user inputs before they reach the model
   Reveal Solution Discussion Select an answer first