The architecture of trust: designing ethical and legal compliance into AI systems
The architecture of trust: designing ethical and legal compliance into AI systems

Summary/Abstract: The deployment of artificial intelligence (AI) in public administration has expanded rapidly, with algorithmic systems now adjudicating welfare eligibility, allocating public resources, and assisting in legal determinations. The prevailing model of AI governance, reactive regulation imposed after deployment, has proven structurally inadequate. High-profile failures such as COMPAS in the United States, SyRI in the Netherlands, and Robodebt in Australia demonstrate that algorithmic harms are not incidental but architectural: the predictable consequence of systems designed without embedded compliance safeguards. This paper advocates a paradigm shift from reactive legal compliance to a proactive “architecture of trust,” in which ethical principles and legal mandates are embedded in the technical architecture of AI systems during the design phase. Drawing on Privacy by Design, Value Sensitive Design, and Legal Protection by Design, the paper investigates how administrative law can function as a foundational engineering blueprint rather than a post hoc compliance checklist. The research addresses a profound interdisciplinary epistemic gap: while engineers require formalized, quantifiable parameters, legal frameworks rely on contextual interpretation, proportionality analysis, and jurisprudential nuance. Translating doctrines such as due process, procedural fairness, and equal protection into executable code, therefore, raises substantial theoretical and practical challenges. The paper proposes a five-pillar compliance-by-design framework: structural fairness assurance, systemic explainability, data provenance and governance, contestability by design, and continuous accountability, and argues that public institutions can engineer AI systems that intrinsically uphold, rather than merely simulate, administrative justice.

• Details
• Contents