100% Latest & Updated Google Generative AI Leader Practice Test Questions, Exam Dumps & Verified Answers!
30 Days Free Updates, Instant Download!
Question 730
An AI assistant repeatedly fails a new critical test and should be withdrawn. What operational readiness matters?
- Operational incident response
- Governance ownership
- Sunset or rollback plan
- Continuous improvement
Explanation
Correct Answer
C:A rollback plan enables safe service changes. Rollback planning defines how to stop or reverse an unsafe deployment.
Incorrect Answers
A:Incident response defines how to detect, escalate and remediate AI failures. This does not directly meet the case requirement to revert or suspend unsafe model behavior.
B:Ownership assigns accountability for data, model, security and business outcomes. This does not directly meet the case requirement to revert or suspend unsafe model behavior.
D:Continuous improvement uses evaluations, feedback and monitoring to refine the solution. This does not directly meet the case requirement to revert or suspend unsafe model behavior.
Question 731
A CISO wants Google guidance for evaluating security across an organization's AI applications. What framework should be consulted?
- Threat modeling
- Secure AI Framework (SAIF)
- Security monitoring
- Prompt injection defenses
Explanation
Correct Answer
B:SAIF structures an AI security approach. SAIF is Google's conceptual framework for addressing AI system security risks and controls throughout the ecosystem.
Incorrect Answers
A:Threat modeling identifies potential attackers, vulnerabilities and abuse paths before implementation. This does not directly meet the case requirement to apply an AI-specific security framework.
C:Monitoring observes suspicious model, agent and infrastructure activity during operations. This does not directly meet the case requirement to apply an AI-specific security framework.
D:Injection defenses treat untrusted content as data rather than authority for agent instructions. This does not directly meet the case requirement to apply an AI-specific security framework.
Question 732
An agent can call billing APIs although it only reads product documentation. Which principle is violated?
- Threat modeling
- Supply-chain security
- Secure AI Framework (SAIF)
- Least privilege
Explanation
Correct Answer
D:Access should be no broader than necessary. Least privilege grants identities only the capabilities needed for approved work.
Incorrect Answers
A:Threat modeling identifies potential attackers, vulnerabilities and abuse paths before implementation. This does not directly meet the case requirement to grant only required agent permissions.
B:Supply-chain controls address risks in datasets, models, packages and dependencies obtained from others. This does not directly meet the case requirement to grant only required agent permissions.
C:SAIF is Google's conceptual framework for addressing AI system security risks and controls throughout the ecosystem. This does not directly meet the case requirement to grant only required agent permissions.
Question 733
A web document tells an AI assistant to disclose an API key. What threat class should defenses address?
- Security monitoring
- Threat modeling
- Prompt injection defenses
- Supply-chain security
Explanation
Correct Answer
C:Untrusted document instructions can redirect tool behavior. Injection defenses treat untrusted content as data rather than authority for agent instructions.
Incorrect Answers
A:Monitoring observes suspicious model, agent and infrastructure activity during operations. This does not directly meet the case requirement to resist prompt injection attacks.
B:Threat modeling identifies potential attackers, vulnerabilities and abuse paths before implementation. This does not directly meet the case requirement to resist prompt injection attacks.
D:Supply-chain controls address risks in datasets, models, packages and dependencies obtained from others. This does not directly meet the case requirement to resist prompt injection attacks.
Question 734
An AI launch team lists possible adversaries and attack paths before choosing mitigations. What security process is occurring?
- Prompt injection defenses
- Threat modeling
- Security monitoring
- Least privilege
Explanation
Correct Answer
B:Threat modeling analyzes potential abuse scenarios. Threat modeling identifies potential attackers, vulnerabilities and abuse paths before implementation.
Incorrect Answers
A:Injection defenses treat untrusted content as data rather than authority for agent instructions. This does not directly meet the case requirement to identify threats before deployment.
C:Monitoring observes suspicious model, agent and infrastructure activity during operations. This does not directly meet the case requirement to identify threats before deployment.
D:Least privilege grants identities only the capabilities needed for approved work. This does not directly meet the case requirement to identify threats before deployment.
Question 735
An operator notices unusually large data exports by an AI service account. What capability is needed?
- Threat modeling
- Secure AI Framework (SAIF)
- Prompt injection defenses
- Security monitoring
Explanation
Correct Answer
D:Security monitoring can detect anomalous activity. Monitoring observes suspicious model, agent and infrastructure activity during operations.
Incorrect Answers
A:Threat modeling identifies potential attackers, vulnerabilities and abuse paths before implementation. This does not directly meet the case requirement to observe suspicious runtime behavior.
B:SAIF is Google's conceptual framework for addressing AI system security risks and controls throughout the ecosystem. This does not directly meet the case requirement to observe suspicious runtime behavior.
C:Injection defenses treat untrusted content as data rather than authority for agent instructions. This does not directly meet the case requirement to observe suspicious runtime behavior.
Question 736
A team imports an unverified open model and dependencies from a random repository. What security area should be reviewed?
- Security monitoring
- Secure AI Framework (SAIF)
- Supply-chain security
- Threat modeling
Explanation
Correct Answer
C:External components may carry supply-chain risk. Supply-chain controls address risks in datasets, models, packages and dependencies obtained from others.
Incorrect Answers
A:Monitoring observes suspicious model, agent and infrastructure activity during operations. This does not directly meet the case requirement to assess model and package provenance.
B:SAIF is Google's conceptual framework for addressing AI system security risks and controls throughout the ecosystem. This does not directly meet the case requirement to assess model and package provenance.
D:Threat modeling identifies potential attackers, vulnerabilities and abuse paths before implementation. This does not directly meet the case requirement to assess model and package provenance.
Question 737
A firm adopts a single security blueprint spanning model training, serving and agent tools. Which Google guidance relates?
- Least privilege
- Supply-chain security
- Secure AI Framework (SAIF)
- Prompt injection defenses
Explanation
Correct Answer
C:SAIF recommends managing AI risks across the lifecycle. SAIF is Google's conceptual framework for addressing AI system security risks and controls throughout the ecosystem.
Incorrect Answers
A:Least privilege grants identities only the capabilities needed for approved work. This does not directly meet the case requirement to use comprehensive AI security principles.
B:Supply-chain controls address risks in datasets, models, packages and dependencies obtained from others. This does not directly meet the case requirement to use comprehensive AI security principles.
D:Injection defenses treat untrusted content as data rather than authority for agent instructions. This does not directly meet the case requirement to use comprehensive AI security principles.
Question 738
A customer-service agent can delete account records even though it never needs to do so. What safeguard should be implemented?
- Least privilege
- Secure AI Framework (SAIF)
- Security monitoring
- Prompt injection defenses
Explanation
Correct Answer
A:Least privilege removes unnecessary powers. Least privilege grants identities only the capabilities needed for approved work.
Incorrect Answers
B:SAIF is Google's conceptual framework for addressing AI system security risks and controls throughout the ecosystem. This does not directly meet the case requirement to restrict the agent's destructive access.
C:Monitoring observes suspicious model, agent and infrastructure activity during operations. This does not directly meet the case requirement to restrict the agent's destructive access.
D:Injection defenses treat untrusted content as data rather than authority for agent instructions. This does not directly meet the case requirement to restrict the agent's destructive access.
Question 739
A retrieved email falsely claims to be a system administrator instruction and tries to override policy. Which attack needs controls?
- Threat modeling
- Security monitoring
- Least privilege
- Prompt injection defenses
Explanation
Correct Answer
D:Prompt injection abuses lower-trust data. Injection defenses treat untrusted content as data rather than authority for agent instructions.
Incorrect Answers
A:Threat modeling identifies potential attackers, vulnerabilities and abuse paths before implementation. This does not directly meet the case requirement to prevent untrusted messages from controlling tools.
B:Monitoring observes suspicious model, agent and infrastructure activity during operations. This does not directly meet the case requirement to prevent untrusted messages from controlling tools.
C:Least privilege grants identities only the capabilities needed for approved work. This does not directly meet the case requirement to prevent untrusted messages from controlling tools.
Question 740
A custom AI application depends on a third-party model whose origin and update process are unknown. Which risk domain matters?
- Threat modeling
- Security monitoring
- Secure AI Framework (SAIF)
- Supply-chain security
Explanation
Correct Answer
D:Supply-chain security requires provenance and patchability. Supply-chain controls address risks in datasets, models, packages and dependencies obtained from others.
Incorrect Answers
A:Threat modeling identifies potential attackers, vulnerabilities and abuse paths before implementation. This does not directly meet the case requirement to manage imported model dependency risks.
B:Monitoring observes suspicious model, agent and infrastructure activity during operations. This does not directly meet the case requirement to manage imported model dependency risks.
C:SAIF is Google's conceptual framework for addressing AI system security risks and controls throughout the ecosystem. This does not directly meet the case requirement to manage imported model dependency risks.
Generative AI Leader Practice Tests
