Risk Assessment
The process of identifying, analyzing, and prioritizing potential harms or failures in AI systems to determine appropriate mitigation strategies.
Définition
A systematic activity that catalogs threats (e.g., bias, privacy breaches, security exploits), evaluates their likelihood and potential impact (financial, reputational, safety), and ranks them by risk score. The output is a risk register that drives mitigation planning. Risk assessments are revisited whenever models, data, or operational contexts change, ensuring that emerging threats are captured throughout the AI lifecycle.
Exemple concret
Before deploying a credit-decision AI, a bank’s risk team maps risks (e.g., false positives denying credit, data leaks), assigns likelihood/impact ratings, and identifies top risks. They then develop targeted controls—enhanced human review for denied applications and encrypted logging—to address the highest-priority items.
.svg)
