Coordinating fragmented global AI governance: Control, Rights, and Safety as a regime complex

Abstract Artificial intelligence governance is often discussed as if it were a single global policy domain. In practice, governments confront a coordination problem across partially overlapping agendas pursued through different instruments and actor configurations. This research note conceptualizes global AI governance as a non-exhaustive regime complex organized around three analytically distinct governance logics: Control, Rights, and Safety. The three are selected because each combines a distinct problem definition, recurrent instrument bundle, actor configuration, and set of portable policy objects. Using an Overton-derived corpus of 198,823 AI-related policy documents and a monthly actor-by-regime panel for 1994–2025, I compare metadata-indexed trajectories, cross-regime overlap, leadership shares, temporal alignment, and OECD-relative post-event salience around three named policy moments. A stratified full-text calibration using 73 documents that were substantively readable by both independent coders, drawn from an initial sample of 100, produced high inter-coder reliability but only partial correspondence between metadata and full-text judgments; correspondence was strongest for Control and weakest for Safety. Within that explicitly documentary scope, Control has the highest metadata prevalence (65.8% of regime-bearing documents) but comparatively unsettled leadership and event-linked uptake. Rights is the clearest IGO-centered regime and shows a short-window post-2019 signal that does not survive stronger interpretations of convergence. Safety has the clearest hybrid public-private configuration. Most metadata-coded regime-bearing documents register one regime, while 13.3% combine two or three, locating candidate interfaces for closer examination. The results imply that coordination, coherence, and integration must be designed differently across regimes rather than pursued through a single omnibus strategy.