Question 1
An AI service is retrained using live feedback without checking whether labels became biased. The concern is on the agenda for a risk management meeting. What recurring safeguard should be prioritized after the initial fix?
Explanation
Correct Answer
Monitor post-update outcomes and trigger investigation on drift. This provides continuing control over the recurring risk that uncontrolled feedback loops can introduce new data and performance risks.
Incorrect Answers
This safeguard addresses a different recurring problem: absent system documentation prevents informed downstream decisions. Here, the risk to control is that uncontrolled feedback loops can introduce new data and performance risks.
This safeguard addresses a different recurring problem: opaque incident handling can delay necessary warnings and remediation. Here, the risk to control is that uncontrolled feedback loops can introduce new data and performance risks.
This safeguard addresses a different recurring problem: one-time testing cannot cover an evolving misuse and attack surface. Here, the risk to control is that uncontrolled feedback loops can introduce new data and performance risks.