A Moral Agency Framework for Legitimate Integration of AI in Bureaucracies

Responsibility doesn't have to disappear with AI in government: The paper challenges the assumption that using AI in bureaucracies inevitably creates "ethics sinks"—situations where responsibility gets lost. Instead, these problems result from specific design choices, meaning they can be prevented through intentional structural decisions in both the technology and the institution.

Three concrete requirements for responsible AI in government: The framework establishes that legitimate AI use requires: (1) maintaining clear human accountability lines, (2) enabling workers to verify AI systems are working correctly, and (3) only deploying AI where it doesn't undermine the bureaucracy's core missions of legitimacy and stewardship.

AI can enhance transparency rather than hide it: Rather than obscuring how government decisions are made, the paper argues AI systems can be designed as tools that increase organizational transparency and accountability—continuing the rationalization that previous waves of digitalization already started in public institutions.

Human verification of AI is essential: The framework emphasizes that workers whose jobs are augmented by AI must be able to verify the systems are functioning properly, placing the burden on institutions to ensure AI remains understandable and auditable rather than treating it as a "black box."

Design choices matter more than the technology itself: The paper's core insight is that problems with AI in government aren't inevitable—they stem from how organizations choose to implement and structure these systems, suggesting that responsible integration is achievable with proper planning.