Harm Assessment
Evaluating potential negative impacts (physical, psychological, societal) of AI systems and defining mitigation strategies.
Définition
A targeted review that categorizes harms (safety, privacy, economic, reputational) specific to the AI application. It uses stakeholder impact mapping, severity-likelihood scoring, and mitigation planning (design changes, guardrails, human oversight). Results feed into risk registers and inform whether to proceed, pause, or redesign the system.
Exemple concret
Before launching an automated-credit-decision AI, a bank conducts a harm assessment: they identify potential economic harms (denial of services), reputational harms (public backlash), and propose mitigations (appeal processes, external review panels). This ensures that decision pathways and user-support channels are in place before rollout.
.svg)
