AI Governance in Social Work: A Triple Mandate-Informed Accountability Model

AI systems are increasingly deployed across public-sector contexts, including social work, informing decisions about risk assessment, resource allocation, and service delivery. Technologies deployed in social work have to operate on its unique characteristics, such as involuntary client engagement, life-altering and often irreversible decisions, relational practice, professional discretion amid moral uncertainty, and working with structurally marginalized populations. Whether existing AI governance frameworks, developed largely for private-sector and routine public-sector contexts, can be effectively applied in social work setting remains under-examined. We identify six operating conditions that prevailing frameworks presuppose and show how each misaligns with social work practice in ways that compromise algorithmic accountability. Drawing on Staub-Bernasconi's triple mandate, we propose a minimum accountability layer organized around three mandate domains (client, organizational, and professional) across key accountability requirements: transparency, contestability, and redress. Our central argument is that effective AI governance must preserve the professional mandate as the independent ethical fulcrum mediating between organizational power and client rights. This work contributes to an emerging conversation about adapting AI governance knowledge to relational, high-stakes service contexts.