European and United States public agencies are increasingly turning enforcement decisions, risk assessments, and eligibility calls over to artificial intelligence and algorithms. The discretion that once belonged to frontline caseworkers doesn’t vanish, but it relocates upstream to choices about training data, feature weights, and decision thresholds. Those choices quietly become de facto policy, yet they seldom appear in any traceable form in the administrative record agencies must defend when challenged. The shift lands differently across the Atlantic. In the United States, Administrative Procedure Act requirements and due-process principles demand explanations that can be contested, but meaningful fixes usually arrive only after someone sues or an oversight body flags serious problems. Europe leans preventive: constitutional proportionality demands upfront justification for rights impacts, data-protection rules require meaningful insight into algorithmic logic and real human safeguards, and newer AI regulations impose mandatory assessments and transparency obligations meant to ensure traceability before deployment (though rollout has been uneven and delayed in places). While much of the literature fixates on bias, ethics, or technical workarounds, this article focuses on the core issue of transparency and accountability. How can long-standing public-law demand for reasoned, reviewable decisions be woven into algorithmic design and procurement so automated outcomes remain scrutable? The comparison reveals Europe’s proactive, rights-centered model versus the U.S.’s reactive, litigation-driven one, yet both converge on essentials: auditable rationales, substantive human oversight, and ongoing monitoring for model drift. The proposed solution is a hybrid approach examining Europe’s preventive tools (impact assessments, deployment registries) paired with America’s corrective strengths (robust contestation rights, court-ordered discovery, independent audits) that are all reinforced through procurement requirements and judicial insistence on legible records. In early 2026, as algorithmic systems proliferate unevenly across government, this isn’t academic speculation, but a practical path to keep administrative power accountable even as discretion burrows deeper into AI code. Keywords: algorithmic governance, administrative discretion, EU AI Act, GDPR Article 22, APA arbitrary-and-capricious review, transparency, accountability, hybrid regulation, proportionality, due process, public administration, opacity, legibility, high-risk AI
Edward Koellner (Sun,) studied this question.