Question 40
Domain 2 — AI Operations, Lifecycle, and Control EnvironmentWhat method MOST effectively enhances an organization’s capability to identify runtime and through-use attacks on artificial intelligence (AI) systems?
Correct answer: C
Explanation
Enhancing AI observability is the best method because it improves visibility into model behavior during operation, which is when runtime and through-use attacks occur. By monitoring inputs, outputs, and system interactions, an organization can detect anomalies and malicious manipulation as they happen.
Why each option is right or wrong
A. Conducting regular audits of access controls
B. Using static code analysis tools
C. Enhancing AI observability
Runtime and through-use attacks occur while the model is operating, so the most effective control is to instrument the AI stack with telemetry that captures inputs, outputs, prompts, intermediate decisions, and downstream actions in real time. In practice, AI observability provides the monitoring and logging needed to spot anomalous behavior, prompt injection, data poisoning effects, or model manipulation as they happen, rather than after the fact.
D. Restricting AI model interactions to internal networks