From Apophatic Interpretability to Accountable Explainability: A Layered Governance Architecture for Structural Opacity in Explainable AI

ADIA Lab Research Paper Series

Authors: Francisco Herrera, João Gama, Salvador García, María José del Jesus, Marcos López del Prado, Luciano Sanchez

Date Published: May 2026

As AI systems grow more complex—higher-dimensional, nonlinear, distributed—demands for transparency increasingly outpace what can be epistemically justified. Opacity in modern AI is not a failure of disclosure but a structural feature of complex models, one that can resist stable human understanding even in fully specified, deterministic systems.

This paper proposes a layered governance architecture for structural opacity in Explainable AI (XAI), spanning four interdependent levels: an ontological account of complexity-driven opacity; apophatic interpretability as disciplined epistemic boundary-setting; accountable explainability as stakeholder-calibrated communication bound by those limits; and institutional mechanisms operationalizing monitoring, auditability, and refusal conditions under acknowledged uncertainty.

Rather than closing opacity as a transparency gap, we argue it should be governed. By reframing interpretability as bounded epistemic warrant and explainability as accountable communication, the architecture grounds trust in enforceable safeguards rather than exhaustive understanding—yielding a systematic framework for governing structural opacity in XAI.