Question 34
IIIA deployed recommendation model begins surfacing harmful content due to a bug in a new feature rollout. What should an AI-specific incident response plan primarily enable?
Correct answer: B
Explanation
An AI-specific incident response plan should support fast containment when a model starts producing harmful outputs, including “rollback or disabling of the affected model or feature.” It should also enable “rapid detection,” stakeholder communication, and “a root-cause analysis to prevent recurrence,” which are standard incident-response functions for limiting harm and preventing repeat failures.
Why each option is right or wrong
A. Immediate deletion of all historical training data.
Deleting training data is not the first containment step and can hinder investigation and remediation.
B. Rapid detection, rollback or disabling of the affected model or feature, communication to stakeholders, and a root-cause analysis to prevent recurrence.
Under an AI incident-response framework, the immediate objective is containment: once a deployed model is generating harmful outputs, the plan should support rapid detection and the ability to disable or roll back the offending model/feature so the harm stops propagating. It should also require timely notification to affected stakeholders and a documented post-incident root-cause analysis, because incident handling is not complete until the failure mode is understood and corrective controls are put in place to prevent recurrence.
C. Automatic migration of the model to a new cloud region.
Changing cloud regions does not address harmful model behavior caused by a bad rollout.
D. Silent suppression of user complaints.
Suppressing complaints hides evidence and delays response instead of reducing user harm.