AIF-C01 Domain 4: Responsible AI

What to understand

Responsible AI is a lifecycle concern. Bias can enter through historical data, sampling, labels, model design, thresholds or the way an application is used. Testing only overall accuracy can hide poor outcomes for a subgroup.

• Fairness asks whether outcomes create unjustified disparities.
• Explainability helps people understand influential factors and model behaviour.
• Transparency includes communicating system purpose, limitations and AI involvement.
• Human oversight is especially important for high-impact or uncertain decisions.

Common AIF-C01 decisions

Choose representative evaluation data, measure relevant subgroups, document limitations and create escalation paths. Guardrails and content filters reduce risk but do not replace governance or human accountability.