Mandatory Institutional Architecture for Public-Sector AI Accountability: A Cross-Jurisdictional Framework Grounded in Administrative Law Version 2.0

Artificial intelligence (AI) is increasingly deployed by states to inform or execute public decisions in areas such as welfare, migration, and law enforcement. Existing legal frameworks impose liability for harms but rarely provide institutional structures to prevent harm or ensure AI-mediated authority remains lawful, reviewable, and accountable. This paper proposes a jurisdiction-neutral accountability framework grounded in administrative and constitutional law. It argues that authority exercised through AI must be governed by mandatory, non-delegable institutional architecture—not merely ethical guidelines or ex post review. The framework establishes four requirements: identifiable political responsibility; designated executive oversight officers; tiered approval and monitoring regimes; and documentary traceability for scrutiny. It is mandatory in function but proportional in form, enabling cross-jurisdictional adaptation while preserving a core of accountability. The paper also addresses implementation challenges such as shadow AI, legacy systems, and autonomous agents, and identifies future research directions. While governance-by-design scholarship (Mökander et al., 2024) and operationalization frameworks (Veale et al., 2018; Kearns & Roth, 2019) advance accountability in AI deployment, they address the AI system as artifact rather than as an instrument of public authority. No existing framework derives mandatory minimum institutional requirements from constitutional and administrative law applicable across jurisdictions. This paper addresses that gap, specifying the institutional conditions under which AI-mediated state action remains lawful, accountable, and democratically legitimate.