What Comes After Harm? Mapping Reparative Actions in AI through Justice Frameworks

Reparation efforts remain largely symbolic: The paper's analysis of 1,060 AI-related incidents reveals that reparative actions are concentrated in early stages like acknowledging harm, with limited progression toward accountability or structural reform—suggesting current responses may not deliver meaningful redress to those affected.

Four key components define adequate reparation: The researchers identify that meaningful reparation requires four elements: acknowledging harm occurred, attributing clear responsibility, providing concrete remedies, and enabling systemic change. Most current AI incidents address only the first one or two of these components.

The field has focused on prevention over repair: While the Responsible AI community has invested heavily in identifying and auditing harms, this paper highlights a significant gap—much less attention has been paid to what actually happens after harm occurs and how to effectively repair it.

Stakeholder involvement in reparations is uneven: The paper's analysis of who initiates and carries out reparative actions shows that responsibility is not equally distributed across companies, regulators, affected communities, and other stakeholders, raising questions about whose voices shape how harms are addressed.

Justice frameworks reveal inadequate responses: By applying established theories of justice to AI incidents, the authors demonstrate that existing reparative approaches fall short of what justice frameworks would consider sufficient, suggesting the need for more comprehensive and accountable remedies.