Identifying Risks to AI Solutions PROFESSIONAL-MACHINE-LEARNING-ENGINEER Practice Questions (Page 5)

Questions 21–25

1. 21application · medium

   A retail company deploys a generative AI application that accepts user reviews and generates product summaries. They have received reports of users submitting reviews that contain malicious instructions designed to alter the summarization behavior. The team wants a managed Google Cloud service that provides content safety filters and prompt injection protection. Which tool should they use?

   • ACloud Data Loss Prevention (DLP)
   • BSecurity Command Center
   • CVertex AI Model Armor
   • DCloud Armor
   Reveal Solution Discussion Select an answer first
2. 22foundation · easy

   Which combination of measures is most effective for preventing unauthorized extraction of data from an AI system?

   • ARegularly updating the AI model to fix vulnerabilities
   • BUsing only strong authentication for all users and service accounts
   • CEncrypting data in transit and at rest, plus monitoring for anomalous access patterns
   • DImplementing network segmentation to isolate the AI system from other resources
   Reveal Solution Discussion Select an answer first
4. 23application · medium

   A media company uses a Vertex AI model to recommend content. The security team notices a spike in prediction requests from a single service account, each requesting a different combination of user IDs. They suspect the account is being used to enumerate the model's training data. Which monitoring signal would most directly confirm this suspicion?

   • AA change in the model's prediction accuracy on a recent batch of validation data.
   • BA sudden increase in the number of distinct user IDs queried by a single service account within a short time window.
   • CAn increase in the number of prediction requests that result in a 404 error.
   • DA decrease in the model's prediction latency, indicating the model is responding faster than usual.
   Reveal Solution Discussion Select an answer first
5. 24foundation · easy

   What is the purpose of using Agent Platform Inference to generate explanations for model predictions?

   • ATo automatically retrain the model
   • BTo encrypt the model's predictions
   • CTo increase the model's prediction speed
   • DTo provide transparency into why a model made a specific prediction
   Reveal Solution Discussion Select an answer first
6. 25foundation · easy

   Besides technical controls, which practice is essential to reduce the risk of users accidentally sharing sensitive data with an LLM?

   • AProviding regular training to users on what constitutes sensitive data and how to handle it safely
   • BUsing a more powerful LLM that can better understand context
   • CDeploying the LLM on-premises to keep data within the organization
   • DIncreasing the model's context window to handle more data
   Reveal Solution Discussion Select an answer first