4.2 Define Secure AI and Its Importance in Protecting AI Systems from Malicious Attacks and Misuse. GENERATIVE-AI-LEADER Practice Questions (Page 5)

Questions 21–25

1. 21application · medium

   A company is deploying a model that was trained on proprietary data. During the deployment phase, they want to ensure that the model is not exfiltrated or tampered with. They also want to monitor for any unusual access patterns that might indicate an attack. Which set of measures should they implement?

   • AUse VPC Service Controls to restrict access to the model, and enable audit logging to track who accesses it.
   • BDisable all logging to reduce the attack surface.
   • CUse a shared service account for all users to simplify access management.
   • DStore the model in a public Cloud Storage bucket for easy access by authorized users.
   Reveal Solution Discussion Select an answer first
2. 22foundation · easy

   What is the primary purpose of Security Command Center in the context of AI workloads?

   • ATo train machine learning models on security data
   • BTo automatically patch all AI model vulnerabilities
   • CTo monitor and manage security risks across AI assets and workloads
   • DTo replace IAM for access control
   Reveal Solution Discussion Select an answer first
4. 23expert · hard

   A company has deployed a gen AI application that uses a large language model. The security team has set up Security Command Center and is receiving a high volume of low-severity findings. They are concerned that the team will suffer from alert fatigue and miss critical threats. They want to improve the effectiveness of their security monitoring. What is the most effective approach?

   • ADisable all low-severity findings to reduce the volume of alerts.
   • BUse Security Command Center's built-in filtering and prioritization features to focus on high-severity findings and tune the detection rules.
   • CIncrease the frequency of scans to generate more data for analysis.
   • DAssign all findings to a single analyst for review.
   Reveal Solution Discussion Select an answer first
5. 24foundation · easy

   What is the primary role of Identity and Access Management (IAM) in protecting AI systems on Google Cloud?

   • AControlling who can access AI resources and what actions they can perform
   • BMonitoring network traffic for malicious activity
   • CAutomatically fixing vulnerabilities in AI models
   • DEncrypting all data used in AI model training
   Reveal Solution Discussion Select an answer first
6. 25expert · hard

   A company is deploying a model that was trained on a public dataset. They are concerned about a specific attack where an attacker crafts inputs that cause the model to misclassify, but the inputs look normal to humans. They want to defend against this at the deployment stage. Which strategy is most aligned with the principles of SAIF?

   • ARestrict access to the model using IAM so only authorized users can send inputs.
   • BRetrain the model on a larger dataset to improve its overall accuracy.
   • CImplement adversarial input filtering and validation before the model processes the input.
   • DUse a more complex model architecture to make it harder for attackers to find adversarial examples.
   Reveal Solution Discussion Select an answer first