From innovation to accountability: Governing AI in public administration
Pedro Robles, Daniel J. MallinsonBackground
Artificial intelligence (AI) is increasingly used in public administration, creating new challenges involving accountability, transparency, explainability, and workforce preparedness. Although state governments are developing AI governance frameworks, it remains unclear whether accountability principles are translated into practical mechanisms for governing AI use.
Objective
This study examines how state AI governance policies operationalize accountability in public administration and identifies remaining gaps in responsibility, oversight, and enforceability.
Methods
A qualitative comparative policy analysis was conducted using publicly available AI governance frameworks from 24 U.S. states identified through March 2026. The documents were analyzed using thematic categories addressing accountability and related governance mechanisms, including transparency, human oversight, auditing, risk management, training, documentation, public notice, appeals, and enforcement. The study distinguished between symbolic accountability, in which accountability is expressed primarily as a governance principle, and operational accountability, in which policies establish identifiable mechanisms for implementation and oversight.
Results
Eighteen of the 24 states explicitly identified accountability as an AI governance principle. However, accountability was frequently expressed symbolically rather than through operational mechanisms. State frameworks more consistently addressed principles such as transparency and risk management while providing less direction regarding enforcement, appeals, independent review, documentation, monitoring, and responsibility for harmful AI-assisted outcomes.
Conclusions
State governments increasingly recognize accountability as an important component of AI governance, but mechanisms for putting accountability into practice remain uneven. Effective public-sector AI governance requires clearer responsibility structures, human oversight, documentation, auditability, review and appeal procedures, and workforce preparation. Without these mechanisms, accountability may remain largely symbolic and responsibility may be shifted toward public-sector workers who are expected to use and justify AI-assisted decisions.