Governance Laundering: A Taxonomy of Failure Modes in AI Compliance Architectures

Enterprise adoption of AI governance tools has accelerated across regulated industries, yet methods for evaluating whether those tools produce verifiable compliance evidence remain underdeveloped. Organizations frequently report that AI governance is “in place,” but the architectural properties required to demonstrate per-decision policy authorization are often absent. This paper introduces a structural taxonomy of governance failure modes—governance laundering, policy theater, and five related families—that arise when governance artifacts are mistaken for governance evidence. We define seven evidence-grade governance requirements (R1–R7) and classify failure modes according to the structural property each violates, including enforcement binding, independence of verification, state completeness, non-delegability, pre-execution authorization, custody continuity, and semantic stability. Building on prior distinctions between visibility, alignment, and authorization layers in AI governance architectures, we formalize the Expanded Anti-Laundering Protocol (EALP), a reproducible seven-test diagnostic framework for determining whether a system produces independently verifiable, per-decision policy evidence. The paper contributes a definitional boundary between governance artifacts and governance evidence, a structured failure-mode taxonomy, a graded evaluation protocol with worked examples, and an architectural susceptibility mapping across common categories of governance products. The framework is vendor-neutral and architecture-focused; it does not propose a new governance design but provides a methodology for evaluating whether existing systems satisfy evidentiary standards required in regulated environments. Implications for procurement, audit, and regulatory oversight are discussed, along with limitations and directions for empirical benchmarking.