Rethinking AI governance for social work through the triple mandate

As artificial intelligence (AI) systems become increasingly integrated into social work practice, questions arise about how they should be evaluated, monitored, and held accountable. Current AI governance frameworks, developed primarily for the private and general public sector contexts, rest on presuppositions that are rarely made explicit: AI deployment is inherently desirable, risks can be decomposed and addressed sequentially, institutions have adequate capacity to implement governance requirements, and affected populations participate voluntarily, with the ability to decline or appeal. Social work’s distinctive conditions (i.e. involuntary client engagement, irreversible high-stakes decisions, relational practice, professional discretion, and service to historically marginalized populations) create a context in which standard AI governance approaches fall short of securing accountability. Rather than proposing a new governance framework, this paper works diagnostically. We surface the conditions that current AI governance frameworks presuppose; trace how their failure in social work compromises transparency, contestability, and redress; and draw on Staub-Bernasconi’s triple mandate to propose a set of guiding questions for what any governance approach must address to remain accountable in this context.