ISACA Certified Artificial Intelligence Auditor (CAIA) — Question 18

Which of the following controls would BEST minimize the risk of bias introduced into AI models from flawed training data?

Answer options

• A. Implementing enhanced access controls over AI model training data
• B. Testing AI models using adversarial methods
• C. Separating AI training and production data
• D. Performing cross-validation of AI models

Correct answer: D

Explanation

Performing cross-validation of AI models helps ensure that the model generalizes well to unseen data and can identify biases introduced during training. The other options, while important for different aspects of model development, do not directly address the mitigation of bias stemming from flawed training data as effectively as cross-validation does.