Which elements constitute a monitoring plan for AI systems?

• A. Ongoing performance metrics only.
• B. Ongoing performance metrics, data drift checks, security events, safety checks, and governance compliance reviews.
• C. Data drift checks and governance compliance reviews only.
• D. Security events and safety checks only.

Answer: (B)

A comprehensive monitoring plan for AI systems spans multiple dimensions: performance, data integrity, security, safety, and governance. Tracking ongoing performance metrics keeps you informed about accuracy, latency, throughput, and error rates, ensuring the system continues to meet service expectations. Data drift checks detect shifts in input data distributions or in the relationships the model learned, signaling when retraining or adjustments may be needed to maintain reliability. Monitoring security events helps identify threats, abnormal access, or attempts to manipulate the model or data, which is essential for protecting confidentiality and integrity. Safety checks verify that outputs remain within safe and acceptable bounds, preventing harmful, biased, or otherwise unsafe results. Governance compliance reviews ensure policies are followed, with proper audit trails, privacy protections, and regulatory and ethical considerations consistently applied. When you consider all these aspects together, you get a robust monitoring plan that addresses performance, reliability, risk, and accountability.