Sample question 1 · Domain IV: Data Privacy and Privacy-Enhancing Technologies (25%)
Arcturus Digital, a large US-based enterprise identity platform serving Fortune 500 clients, operates an SSO system that incorporates an AI model for continuous authentication risk scoring. The model runs inference at the edge — on client-managed endpoint agents installed on employee workstations — to evaluate behavioral signals (keystroke dynamics, mouse movement entropy) without transmitting raw biometric data to central servers. The identity team is preparing a privacy governance report for the company's Chief Privacy Officer and enterprise clients. A key stakeholder concern is whether edge inference on biometric-adjacent behavioral data constitutes processing of sensitive data, and whether the current architecture adequately controls for model leakage from compromised endpoints. As the privacy technologist preparing the governance report for Arcturus Digital's CPO and enterprise clients, which combination of controls most accurately characterizes a privacy-by-design architecture for edge AI inference on behavioral biometric signals, and should be recommended to address the model leakage risk?
- A. Classify behavioral biometric inference as non-sensitive processing because raw biometric templates are never transmitted, and rely on endpoint antivirus controls to prevent model extraction from compromised devices.
- B. Treat edge inference outputs as sensitive given their biometric-adjacent nature, implement on-device model encryption with secure enclave execution to protect model weights, and apply output quantization to limit the precision of risk scores transmitted to central systems.
- C. Transmit all raw behavioral signals to a centralized secure server for inference processing, eliminating edge model leakage risk by removing model artifacts from endpoints entirely.
- D. Require enterprise clients to sign data processing agreements acknowledging that edge inference is their responsibility, and document the risk transfer in the governance report as the primary mitigation for model leakage.
View answer & explanation
Correct answer: B. Treat edge inference outputs as sensitive given their biometric-adjacent nature, implement on-device model encryption with secure enclave execution to protect model weights, and apply output quantization to limit the precision of risk scores transmitted to central systems.
[Domain IV: Data Privacy and Privacy-Enhancing Technologies (25%), Competency IV.A: PETs and Emerging Tech - Area 1 - Model training data] Core Concept: IV.A.2 Evaluate privacy controls for AI, IoT, and cloud architectures. centers on model training data as defined in the official (ISC)² exam outline. Correct Answer (B): Treat edge inference outputs as sensitive given their biometric-adjacent nature, implement on-device model encryption with secure enclave execution to protect model weights, and apply output quantization to limit the precision of risk scores transmitted to central systems. Why the other options are incorrect: Option A: The absence of raw template transmission does not eliminate sensitivity; behavioral biometric inference outputs can still be used to re-identify or profile individuals, and antivirus controls do not address sophisticated model extraction techniques such as side-channel or API-based attacks. Option C: Centralizing inference eliminates the edge architecture's core privacy benefit — keeping sensitive behavioral signals on-device — and introduces new transmission and centralized processing risks that may be disproportionate to the model leakage concern. Option D: Contractual risk transfer to enterprise clients is a legal mechanism, not a technical privacy control; it does not mitigate model leakage risk and misrepresents the privacy technologist's responsibility to implement architectural safeguards rather than simply document liability allocation.