ISACA Certified AI Security Manager — Question 34

Which of the following is the MOST effective way to prevent a model inversion attack?

Answer options

• A. Monitor model output for anomalous patterns.
• B. Utilize data pseudonymization techniques.
• C. Implement differential privacy during model training.
• D. Ensure data minimization best practices.

Correct answer: C

Explanation

Implementing differential privacy during model training effectively adds noise to the output, making it difficult for attackers to reconstruct the original data. The other options, while useful for general security practices, do not specifically address the vulnerability posed by model inversion attacks as directly as differential privacy does.