An Overview of the Risk-based Model of AI Governance

Risk-based AI governance may inadvertently reproduce existing inequalities: The paper argues that how risks are constructed and calculated in current AI regulations like the EU AI Act can embed and perpetuate existing societal inequalities, rather than mitigating them as intended.

"Risk" and "harm" are not the same thing and should be treated differently: The paper distinguishes between these concepts, noting that risk assessments often reflect whose interests regulators consider important, which can obscure actual harms experienced by marginalized groups.

Current AI governance frameworks embed hidden value judgments: The paper contends that risk categorizations in AI policy contain inherent assumptions about what matters and to whom, and these should be openly scrutinized rather than treated as objective technical determinations.

Multiple regulatory approaches may work better than risk-based models alone: The paper suggests that combining risk-based governance with other regulatory tools and "responsive" regulation—that adapts to real-world outcomes—could improve AI governance frameworks.

Existing risk governance scholarship offers practical solutions: Rather than starting from scratch, policymakers can draw on decades of risk regulation research from other fields to strengthen and refine how AI is governed.