4.2 Define Secure AI and Its Importance in Protecting AI Systems from Malicious Attacks and Misuse. GENERATIVE-AI-LEADER Practice Questions (Page 2)

Questions 6–10

1. 6expert · hard

   A company has a gen AI model that is experiencing a slow degradation in performance over several weeks. The security team suspects that the model is being subtly manipulated, but the changes are too gradual to trigger standard alerts. They want to detect this slow-burn attack. Which monitoring strategy is most effective?

   • AIncrease the frequency of standard alerts to catch any small changes immediately.
   • BSet up long-term trend analysis on model performance metrics and input data distributions, with alerts on gradual deviations from the baseline.
   • CManually review all model outputs on a weekly basis.
   • DDisable monitoring for the model to avoid false positives.
   Reveal Solution Discussion Select an answer first
2. 7application · medium

   A healthcare startup is building a medical image classifier. During the data collection phase, they scrape publicly available images from social media without verifying the source or the integrity of the files. A security reviewer warns that this practice introduces a specific type of risk that will propagate through the entire ML lifecycle. What is the most immediate security concern introduced at this stage?

   • AThe model may be poisoned with mislabeled or maliciously crafted images, degrading accuracy and potentially enabling targeted attacks after deployment.
   • BThe training pipeline will fail because social media images are not in a supported format for the ML framework.
   • CThe model will be overfit to the training data, causing poor generalization on real-world medical images.
   • DThe model will be unable to meet HIPAA compliance because the data was not collected under a business associate agreement.
   Reveal Solution Discussion Select an answer first
4. 8application · medium

   A company has a gen AI application that is experiencing unexpected behavior. The security team suspects a prompt injection attack. They need to detect and respond to this threat in real time. Which combination of tools would best enable this?

   • ACloud Functions to process prompts and Cloud Storage to store them.
   • BCloud Armor to block malicious IP addresses and Cloud CDN to cache responses.
   • CCloud Logging to capture prompts and responses, and Cloud Monitoring to set up alerts on anomalous patterns.
   • DCloud IAM to restrict access and Cloud KMS to encrypt data.
   Reveal Solution Discussion Select an answer first
5. 9foundation · easy

   What is the purpose of using Cloud Monitoring for AI workloads?

   • ATo track performance metrics and set alerts for abnormal behavior that could indicate a security issue
   • BTo manage IAM roles for AI users
   • CTo store training datasets securely
   • DTo deploy AI models to production
   Reveal Solution Discussion Select an answer first
6. 10expert · hard

   A large enterprise is migrating its AI workloads to Google Cloud. They have a strict requirement that all AI models and data must be protected from both external attacks and internal misuse. They also need to demonstrate compliance to auditors. They are considering a multi-layered approach. Which combination of Google Cloud security features best addresses these requirements?

   • AEncryption of data at rest and in transit only.
   • BRelying on the cloud provider's default security settings without additional configuration.
   • CSecure-by-design infrastructure, IAM with least privilege, Security Command Center, and audit logging.
   • DA single firewall and a strong password policy.
   Reveal Solution Discussion Select an answer first