Question 440
The risk owner says a control improvement is complete: A firm plans to use a generative AI model for eligibility decisions without evaluating explainability or bias. What outcome would most clearly demonstrate its effectiveness?
Explanation
Correct Answer
Decisions are tested for fairness, traceability and approved human oversight. This demonstrates the intended business or control result, rather than just task completion. The underlying response was: Conduct an AI-specific risk assessment and implement monitoring and accountable review processes. The evidence should relate to: Review regulated decisions, training data limitations, model governance and human appeal rights.
Incorrect Answers
A simulated unsafe route command is stopped before entering occupied zones. That would matter when a robotics vendor introduces autonomous warehouse vehicles that can modify routes without supervisor approval. The relevant distinction in this case is: Decisions are tested for fairness, traceability and approved human oversight.
Critical services continue working while targeted cryptographic controls are upgraded. That would matter when a company adopts post-quantum cryptography everywhere immediately without testing legacy interoperability. It does not address this case's requirement: A firm plans to use a generative AI model for eligibility decisions without evaluating explainability or bias.
A newly disclosed dependency weakness maps to affected releases and accountable response actions. That would matter when a project uses open-source components but has no process for tracking transitive dependencies and vulnerability notices. That separate mechanism does not verify the needed risk management outcome in this case.