When AI takes control, who takes responsibility? Structural analysis of agentic AI risks in supply chains

Purpose Agentic AI systems execute procurement decisions, reroute shipments and renegotiate supplier contracts at speeds that preclude real-time human review. The existing literature offers limited analytical attention to the risks that arise when decision authority is delegated to agents. This paper develops a risk taxonomy for agentic AI in supply chains and traces the hierarchical dependencies and influence magnitudes among the identified risks. Design/methodology/approach A systematic review of literature, analyzed through theories–characteristics–contexts–methods framework, produces the risk taxonomy. A hybrid modified total interpretive structural modeling and fuzzy decision-making trial and evaluation laboratory methodology is then applied to structured expert judgments to establish the causal hierarchy and quantify inter-risk influence. Findings Three risks, namely workforce skill gap, data quality and reliability and ethical misalignment occupy the deepest level of the causal hierarchy. Lack of accountability and governance and data privacy emerge as terminal outcomes. All three analytical methods converge on a conclusion: agentic AI risks originate at the institutional layer, with technical risks acting as transmission mechanisms. Practical implications The findings indicate that effective governance requires that workforce capability, data integrity and objective-function alignment are addressed before deployment. The stratified agentic AI governance framework is proposed that translates hierarchical structure into a sequenced intervention architecture, providing practitioners with governance priorities organized by risk-propagation logic. Originality/value This paper contributes a systematically derived risk taxonomy and a causally weighted governance map for agentic AI risks in supply chains. Five propositions are derived from the causal structure and are advanced as a testable research agenda.