Administrative Law's Fourth Settlement: AI and the Capability-Accountability Trap

TL;DR

This paper proposes a new administrative law framework leveraging AI to enhance transparency and oversight, addressing the capability-accountability trap that has historically hindered democratic control over complex agencies.

Contribution

It introduces three doctrinal innovations—Model and System Dossier, material-model-change trigger, and deference to audit—to enable AI-assisted transparent oversight in administrative law.

Findings

AI can translate technical complexity into accessible oversight information.

The proposed framework aims to balance agency capability with democratic accountability.

These innovations facilitate auditable and comprehensible AI decision-making in government.

Abstract

Since 1887, administrative law has navigated a "capability-accountability trap": technological change forces government to become more sophisticated, but sophistication renders agencies opaque to generalist overseers like the courts and Congress. The law's response--substituting procedural review for substantive oversight--has produced a sedimentary accretion of requirements that ossify capacity without ensuring democratic control. This Article argues that the Supreme Court's post-Loper Bright retrenchment is best understood as an effort to shrink administration back to comprehensible size in response to this complexification. But reducing complexity in this way sacrifices capability precisely when climate change, pandemics, and AI risks demand more sophisticated governance. AI offers a different path. Unlike many prior administrative technologies that increased opacity alongside capacity, AI can help build "scrutability" in government, translating technical complexity into accessible terms, surfacing the assumptions that matter for oversight, and enabling substantive verification of agency reasoning. This Article proposes three doctrinal innovations within administrative law to realize this potential: a Model and System Dossier (documenting model purpose, evaluation, monitoring, and versioning) extending the administrative record to AI decision-making; a material-model-change trigger specifying when AI updates require new process; and a "deference to audit" standard that rewards agencies for auditable evaluation of their AI tools. The result is a framework for what this Article calls the "Fourth Settlement," administrative law that escapes the capability-accountability trap by preserving capability while restoring comprehensible oversight of administration.