Ankesh Chandaria (University of Cambridge; Leverhulme Centre for the Future of Intelligence) has posted When Constitutional AI Becomes Public Infrastructure: A Threshold for Democratic Legitimacy on SSRN. Here is the abstract:
Constitutional AI (CAI) governs large-language-model behaviour through privately authored normative frameworks. As these systems become embedded across public administration, they raise democratic-legitimacy questions not adequately captured by existing critiques of opacity, community disconnect, or public-authority publicity. This paper argues that such frameworks create a legitimacy deficit once they cross the Infrastructure Threshold: the point at which a CAI system ceases to be a product users may decline and becomes a condition of access to entitlements they cannot meaningfully avoid. The threshold is reached when three conditions converge: institutional embedding, practical indispensability, and normative gatekeeping. Above it, democratic legitimacy must attach to the authorship of governing norms, not merely to their outputs. The argument is anticipatory, addressing a governance model being institutionalised through public-sector pilots, procurement, and AI-use inventories before it hardens into settled infrastructure. Drawing on Rousseau, Habermas, and Pettit, the paper identifies democratic self-authorship as the relevant legitimating condition. The resulting deficit persists even where systems are transparent, outputs are reasonable, and procurement frameworks attempt to constrain them. Procurement reform may offer a tractable near-term intervention, but cannot fully satisfy self-authorship.
Recommended!
To receive new posts from Legal Theory Blog by email, get a free subscription to Legal Theory Stack.
Lawrence Solum