What are Key Risk Indicators (KRIs) in AI security?

• A. Metrics that measure risk and bias mitigation, such as disparity and fairness metrics.
• B. Purely financial metrics like revenue per model.
• C. Only data throughput.
• D. Only hardware failure rates.

Answer: (A)

KRIs in AI security are metrics used to monitor potential risks in AI systems, focusing on outcomes that could cause harm or noncompliance rather than just raw performance. These indicators track issues like bias, fairness, privacy exposure, data quality, and security vulnerabilities so you can detect and mitigate risk early. For example, disparity metrics reveal how differently the model treats different groups, while fairness metrics quantify whether outcomes are equitable. These signals help govern AI deployments by guiding thresholds, alerts, and remediation steps.

That’s why the option describing metrics that measure risk and bias mitigation, such as disparity and fairness metrics, is the best fit. It centers on monitoring risk in AI systems and actively addressing bias, which is central to AI security governance. The other choices miss the risk-focused aspect: pure financial metrics don’t capture safety or fairness concerns, throughput is an operational measure, and hardware failure rates address reliability rather than AI-specific risk.