AI Security Architecture and Design AAISM Practice Questions (Page 2)

Questions 6–10

1. 6foundation · medium

   What is the primary purpose of an AI security architecture review?

   • ATo optimize the model's accuracy and performance
   • BTo identify security gaps and improvements against best practices and regulatory requirements
   • CTo automate the deployment of AI models
   • DTo reduce the cost of AI infrastructure
   Reveal Solution Discussion Select an answer first
2. 7application · medium

   A healthcare organization is developing an AI model to predict patient readmission risks. During threat modeling, the team identifies that an attacker could subtly alter the training data to cause the model to make incorrect predictions for certain patient groups. Which threat is being identified, and what is the most appropriate control to mitigate it?

   • APrompt injection; implement input sanitization on all data fed to the model.
   • BData poisoning; implement strict data provenance and integrity checks on training data.
   • CAdversarial attack; implement adversarial training to make the model robust.
   • DModel inversion; implement differential privacy on the model outputs.
   Reveal Solution Discussion Select an answer first
4. 8application · medium

   A company is reviewing its AI architecture for compliance with GDPR. The AI system processes personal data of EU citizens. The review finds that the data is stored in a data center in the EU, but the model training is performed on a cloud platform in the US. What is the primary compliance concern?

   • AData availability: the training platform may not be available.
   • BData sovereignty: the US government could access the data.
   • CData integrity: the training process may alter the data.
   • DData residency: the personal data may be transferred outside the EU during training.
   Reveal Solution Discussion Select an answer first
5. 9foundation · medium

   Which security control for AI models is used to detect and alert on unusual patterns of API calls that may indicate unauthorized use or model extraction?

   • AEncryption of model weights
   • BAccess control lists
   • CInput validation
   • DMonitoring for unauthorized use
   Reveal Solution Discussion Select an answer first
6. 10foundation · medium

   Which infrastructure security measure is used to isolate AI workloads from each other and from other parts of the network to limit the blast radius of a compromise?

   • ALog aggregation
   • BPatch management
   • CNetwork segmentation
   • DData encryption
   Reveal Solution Discussion Select an answer first