Sample question 1 · AI Governance and Program Management (31%)
Veltronix Precision Components, a mid-sized German automotive parts manufacturer, operates computer-vision (CV) models on its stamping lines to detect surface defects in real time. During a routine shift handover, a line technician named Markus notices that the CV system has been flagging an unusually high number of false positives for the past 90 minutes, causing repeated line stoppages. Markus assumes it is a camera calibration issue and manually overrides the AI rejection flags to keep production running. He does not log the override or notify the AI security team. Two hours later, the quality assurance supervisor discovers that 4,200 potentially defective parts have shipped to a Tier-1 customer. The AI security program lead is now reviewing the incident and finds that Veltronix has no formal AI-specific acceptable use guideline addressing operator override authority or escalation triggers for anomalous model behavior. As the AI security program lead at Veltronix, which action should you prioritize FIRST to prevent a recurrence of this type of incident?
- A. Immediately retrain the CV model using the flagged false-positive images to improve its accuracy before issuing any policy update.
- B. Establish an AI acceptable use guideline that defines operator override authority limits, mandatory logging requirements, and explicit escalation triggers when model behavior deviates from baseline thresholds.
- C. Require all production line technicians to complete a general cybersecurity awareness course covering phishing and password hygiene within the next 30 days.
- D. Commission a third-party audit of the CV model's training data to determine whether data poisoning caused the false-positive surge before taking any policy action.
View answer & explanation
Correct answer: B. Establish an AI acceptable use guideline that defines operator override authority limits, mandatory logging requirements, and explicit escalation triggers when model behavior deviates from baseline thresholds.
[Domain 1 - AI Governance and Program Management, Subtopic 1B - AI-Related Strategies, Policies, and Procedures - Area 1 - AI-Related Strategies, Policies, and Procedures] Core Concept: AI-specific acceptable use guidelines must define the boundaries of human interaction with AI systems, including who may override model outputs, under what conditions, and what escalation steps are mandatory. Without these controls, operators act on personal judgment, creating unlogged, ungoverned interventions that bypass the AI system's intended safety function and obscure security or quality incidents. Correct Answer (B): The root cause of this incident is the absence of a policy governing operator override authority and escalation obligations. Establishing an acceptable use guideline that codifies override limits, mandatory logging, and escalation triggers directly addresses the governance gap and prevents future unilateral, unrecorded interventions. Correct Answer (B): Correct Answer (B): Correct Answer (B): This is the foundational control that all subsequent technical or training measures depend upon. Why the other options are incorrect: Why the other options are incorrect: Why the other options are incorrect: Why the other options are incorrect: Option A: : : : Retraining the model addresses a potential technical symptom but does not resolve the governance failure — operators would still have no policy guidance on when and how overrides are permitted, leaving the same behavioral gap in place. Option C: : : : A generic cybersecurity awareness course covering phishing and password hygiene is irrelevant to the specific AI-human interaction risk exposed here; AI-specific acceptable use training tied to model override procedures is what is required. Option D: : : : Commissioning a data-poisoning audit may be warranted as a parallel investigation, but delaying policy action until the audit concludes leaves the override gap open and does not address the immediate governance deficiency that caused the business impact.