Existing governance institutions fail on all four core functions when applied to AI: monitoring is defeated by private labs and weights-level opacity, standard-setting lacks technical grounding, enforcement has no jurisdiction over global compute, and legitimacy erodes as governed parties outpace governors. New institutions must be designed as an engineering problem, not adapted from nuclear or spectrum governance models.
Strongest rival: Existing institutions can be reformed and extended -- the EU AI Act, NIST frameworks, and voluntary commitments show that existing governance structures are adapting, not failing
Claim
From 'New Institutions for Powerful AI' by The Editors (Pax Machina Magazine, August 2026, paxmachina.ai/welcome-to-pax-machina). Lays out six themes for institutional design: (1) concentration vs distribution of power, (2) AI economy and social contract, (3) governance/oversight/alignment/control, (4) defense and security, (5) information/truth/epistemics, (6) human flourishing. Frames the AI transition as compressed into years rather than generations, making institutional adaptation urgent. The editorial board includes Seth Lazar, Saffron Huang, Iason Gabriel, Dean Ball -- serious AI governance researchers.