Objective 4.2 Secure a Pipeline (includes Maintenance) AIP-210 Practice Questions (Page 2)

Questions 6–10

1. 6application · medium

   An e-commerce company trains a recommendation model and stores the serialized model file (e.g., .pkl) in Azure Blob Storage. The ML team is concerned about model tampering, where an attacker with write access to the storage account could replace the model with a malicious version. They want to ensure that the model loaded for inference is exactly the one that was trained and validated. What should they implement?

   • AStore the model file in a container with a long, random name and rely on the container's URL being secret.
   • BCopy the model file to a second storage account and compare the file sizes before each inference.
   • CCompress the model file with a password-protected ZIP archive and store the password in the pipeline's environment variables.
   • DCompute a SHA-256 hash of the model file after training, store the hash in a separate secure location, and verify the hash before loading the model for inference.
   Reveal Solution Discussion Select an answer first
2. 7application · medium

   A logistics company's ML pipeline runs on a schedule and uses a service principal to authenticate to Azure resources. The security team suspects that the service principal's credentials may have been compromised and wants to detect any unauthorized use of the service principal in the pipeline. What should they configure to detect this?

   • ADisable the service principal and create a new one, then update the pipeline to use the new credentials.
   • BEnable Azure AD sign-in logs and create an alert for sign-ins from the service principal that occur outside the scheduled pipeline run times or from unexpected IP addresses.
   • CMonitor the pipeline's compute metrics and alert if CPU usage is above 50%.
   • DReview the Azure Activity Log for 'write' operations on the service principal's key vault.
   Reveal Solution Discussion Select an answer first
4. 8expert · hard

   A financial technology company has an ML pipeline with multiple stages: data ingestion, feature engineering, model training, and model deployment. The pipeline uses a shared service principal for all stages. A recent security review found that the service principal has more permissions than needed, and the review also noted that there is no way to determine which stage of the pipeline accessed a particular data asset. The company wants to reduce the blast radius of a compromised credential and improve auditability. What is the most appropriate action?

   • ACreate a separate service principal for each pipeline stage, assign each only the permissions required for that stage, and enable diagnostic logging on the data assets to capture which principal accessed them.
   • BKeep the single service principal but rotate its credentials monthly and enable Azure AD sign-in logs.
   • CUse the same service principal but add a custom header to each pipeline stage's API calls to identify the stage.
   • DReplace the service principal with a managed identity for the entire pipeline and grant it Contributor access to the resource group.
   Reveal Solution Discussion Select an answer first
5. 9expert · hard

   A large enterprise runs a critical ML scoring service on Azure Kubernetes Service (AKS). The cluster is running an older Kubernetes version that is approaching end-of-life, and the node images have known vulnerabilities. The ML team must upgrade the cluster and the model's base image, but the service has a strict uptime SLA and cannot tolerate downtime. The team also has a limited maintenance window on weekends. What is the most appropriate upgrade strategy?

   • AUpgrade the Kubernetes version first, then immediately delete and recreate all pods to force the new image to be pulled.
   • BPerform a rolling upgrade of the AKS cluster to the new Kubernetes version, then rebuild the model image from a patched base image and deploy it using a blue/green strategy with automated health checks before traffic switchover.
   • CPatch the node images manually via SSH and update the image tag in the deployment manifest without rebuilding the image.
   • DRecreate the entire AKS cluster with the new version and redeploy the model using the existing image, scheduling the recreation during the maintenance window.
   Reveal Solution Discussion Select an answer first
6. 10foundation · easy

   Which access control principle ensures that users and services are granted only the minimum permissions necessary to perform their tasks?

   • AFail-safe defaults
   • BLeast privilege
   • CDefense in depth
   • DSeparation of duties
   Reveal Solution Discussion Select an answer first