Auditabilidad Organizacional de Sistemas de Inteligencia Artificial: Especificación Conceptual de la Capacidad Probatoria y sus Dimensiones

Abstract

This essay developed a conceptual specification of organizational auditability in artificial intelligence (AI) systems as an evidential capability required to support auditing and assurance over AI governance. The analysis distinguished three related but non-equivalent functions: governance allocated responsibilities, identified risks, and operated controls; auditing required sufficient and reliable evidence about those arrangements; and assurance involved an independent conclusion within a defined scope. The proposal organized 24 core concepts across seven logical levels, distinguished intrinsic from accessible auditability, and operationalized the construct through eight dimensions: identifiability, attributability, temporal traceability, integrity, recoverability, linkability, authorized accessibility, and retention. It also articulated a common thread linking governance, risk, control, evidence, auditability, and assurance, reframed the Control × Auditability matrix, and examined the architecture through three synthetic scenarios. The conceptual stress test did not constitute empirical validation; it was used to identify contradictions and inferential boundaries. The essay concluded that formally well-designed governance could remain weakly assurable when it failed to generate verifiable evidence. Auditability was therefore framed as a socio-technical property that should be designed from the outset and was presented as a candidate for subsequent expert and empirical validation.