An IS auditor finds that an AI model's outputs are not being reviewed. Which of the following would BEST address this risk?
Answer(s): B
A validation process for AI decisions ensures that model outputs are consistently reviewed for accuracy, appropriateness, and alignment with business and compliance requirements, directly mitigating the risk of unverified or erroneous outputs.
Which of the following techniques would be MOST effective as part of incident management procedures for a prompt injection attack?
Answer(s): C
Input validation that sanitizes or filters malicious prompts is the most effective way to mitigate prompt injection attacks because it prevents harmful or manipulative input from ever reaching the model, reducing the likelihood of compromised outputs.
Which of the following BEST ensures representativeness in AI systems when assessing training data periodically?
Ensuring that training data remains relevant and aligned with evolving real-world conditions is essential for maintaining representativeness, allowing the AI system to generalize accurately and perform reliably over time.
Which of the following are the MOST appropriate stages in the AI life cycle for evaluating edge cases?
Answer(s): D
Evaluating edge cases is most appropriate during the test and verify stage, where model behavior under unusual or extreme conditions can be assessed before deployment to ensure safety and reliability.
An IS auditor identifies that an AI model occasionally invents nonexistent medical test results. Which of the following recommendations would BEST mitigate this risk?
Answer(s): A
Decreasing top-p sampling reduces the randomness of token selection by narrowing the model’s probability distribution to the most likely outputs, which effectively lowers the likelihood of hallucinations such as fabricated medical test results.
Post your Comments and Discuss ISACA AAIA exam prep with other Community members:
Our website is free, but we have to fight against AI bots and content theft. We're sorry for the inconvenience caused by these security measures. You can access the rest of the AAIA content, but please register or login to continue.