Sources of Risk in AI Systems Revisited: FCAIR-O, a Faceted Causal Framework and Lightweight Ontology for Risk Management and Assurance

Abstract Artificial-intelligence risk frameworks often combine causal sources, vulnerabilities, failure modes, harms, desired properties, controls, evidence, and legal classifications within the same lists. This role conflation obscures mechanisms, intervention points, and responsibility. As the conceptual successor to Steimers and Schneider (2022), this article retains intended purpose, reasonably foreseeable misuse, the protective-measure hierarchy, residual risk including control-induced risk, and post-market observation. It reassesses their eight-category taxonomy through a critical integrative conceptual synthesis supported by an access-limited evidence map, best-fit framework synthesis, legal and standards analysis, and an explicit design audit trail covering material available through 25 July 2026. The resulting Faceted Causal AI Risk and Assurance Ontology (FCAIR-O v1.2.1) is a conceptual framework implemented as a lightweight ontology and controlled concept scheme. It comprises a twelve-node causal and assurance backbone, eight orthogonal facets, twelve parent causal-source classes, and 74 operational causal leaves within 221 versioned concept identifiers. Five design-oriented theoretical propositions require ontological role separation, mechanistic traceability, multi-level causality, control-evidence coupling, and distinct normative and epistemic boundaries. Fairness, privacy, transparency, automation, complexity, security, hardware, and technological maturity are retained where useful but reassigned to causal, harm, property, control, evidence, or normative roles. A documentary reconstruction of the Uber Tempe fatality and additional cross-sector cases illustrate the framework without constituting empirical validation. FCAIR-O is offered as a revisable academic construct for risk-control-evidence records, assurance arguments, case analysis, and legal or sectoral overlays.