Governed Intelligence Architecture for Institutional AI

Abstract

AI adoption has outpaced the epistemic infrastructure required to make institutional AI reliable. In regulated organizations, AI systems do not act on knowledge in the abstract — they act on claims whose provenance, scope, freshness, coherence, and authority determine whether outputs are usable, auditable, and safe. This paper introduces epistemic immunity: a governed defense architecture for protecting institutional knowledge substrates from six systemic failure modes — pollution, staleness, fragmentation, amnesia, cascade failure, and structural distortion.

The paper distinguishes knowledge from intelligence. Knowledge is validated, contextual, and relational understanding distributed across people, systems, records, and practices. Intelligence is the governed capacity to mobilize that knowledge into reliable judgment and action under changing conditions. Governed intelligence therefore requires more than AI capability: it requires knowledge infrastructure that can admit, validate, scope, demote, preserve, and renew claims over time.

Epistemic immunity is operationalized through the Governed Intelligence Lifecycle — Ingest, Consolidate, Curate, Expand, and Apply — and specified through a three-scope capability architecture covering claim-level controls, graph-level integrity, and delivery-level access, explanation, and agent action. The framework draws on distributed cognition, sensemaking, autopoiesis, dissipative structures, and the adjacent possible to argue that institutional knowledge is not a static asset but a continuously regenerated socio-technical system.

Applied to financial services, the paper shows how regulated institutions can move beyond retrieval-oriented AI toward governed intelligence systems that preserve provenance, manage decay, surface contradictions, constrain agent action, and compound learning without accumulating epistemic debt. The paper develops a conceptual architecture grounded in distributed cognition and knowledge governance traditions, operationalized as an implementation methodology, and applied to the regulatory and operational context of financial services.

This is Paper D of a five-paper programme: Paper A (Reichhart and Gelas, 2026a) diagnoses dynamics blindness as the failure mechanism; Paper B (Gelas and Reichhart, 2026b) specifies the architectural resolution — a tripartite structure of state, dynamics, and agency layers; Paper C (Reichhart and Gelas, 2026c) provides theoretical foundations from ten independent traditions; this paper operationalizes the methodology; Paper E (Reichhart and Gelas, 2026e) identifies the conditions under which the architecture produces governed initiative rather than mere autonomy.

Keywords: AI Governance, Epistemic Immunity, Knowledge Graphs, Institutional Intelligence, Epistemic Operational Risk, Path Dependency, Neuro-Symbolic Reasoning, Regulated Industries