Evaluating potential negative impacts (physical, psychological, societal) of AI systems and defining mitigation strategies.
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.
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.
