What should a postmortem review of an AI incident focus on?

• A. Identifying areas for improvement in data preprocessing, security controls, and adversarial testing.
• B. Assigning blame and punishment.
• C. Reducing monitoring controls.
• D. Reducing data quality.

Answer: (A)

Postmortems of AI incidents should center on identifying concrete improvements to prevent recurrence. This means looking at how data handling and preprocessing contributed to the issue, so data quality, labeling, and data pipeline robustness can be strengthened. It also involves evaluating security controls to uncover gaps in access management, model protection, and governance that could let similar problems occur again. And it includes reviewing adversarial testing and resilience measures to understand how the system behaves under targeted or unexpected inputs, enabling stronger defenses and quicker containment in the future. Focusing on these areas turns the incident into a actionable plan for improvement, rather than assigning blame or rolling back safeguards. The other options miss the mark: blaming people or processes, reducing monitoring, or degrading data quality would only increase risk and do not help prevent future incidents.