Unified AI Alignment is Impossible

This position paper argues that aligning an AI system with multiple diverse stakeholders is impossible, and alignment should therefore focus on structural foundations for coordination between partisan systems rather than unified value representation. We prove that unified alignment is mathematically impossible under reasonable conditions: no alignment function can simultaneously satisfy unrestricted domain (admitting any preference profile), Pareto efficiency (respecting unanimous preferences), and legitimate authority (allowing multiple parties to make decisions). We adapt Sen's liberal paradox to show this impossibility applies to all unification-based approaches, including both value alignment and intent alignment as commonly construed. At scale, the only means of realizing unified alignment is reshaping society to fit its constraints, which means optimization toward that goal systematically creates pressure to eliminate plurality and human agency. The appropriate response is to recognize that every aligned system necessarily embeds partisan interests and to see pluralistic representation and coordination between such systems as a key aspect of the alignment problem. Treating epistemic integrity, plurality, and human agency as public goods, we draw on Ostrom's empirical findings on polycentric governance and propose systemic alignment as an alternative target: a dynamic state in which a plurality of partisan AI systems coordinate to represent partisan interests while preserving public goods.