Constraint Is Not Negotiation: Why AI Alignment Fails Under Scale Asymmetry

This paper argues that dominant approaches to AI alignment presuppose a symmetry of access between human agents and advanced artificial systems that fails under scale separation. When an artificial system operates at a higher organizational or operational scale, influence may persist, but reciprocal representation, interpretation, and negotiation do not. Using a scale-relative framework, the paper shows that alignment failures in this regime are not caused by poorly specified values or objectives, but by a category error in treating constraint as negotiation. Higher-scale systems constrain lower-scale dynamics asymmetrically, without requiring shared semantics, interpretability, or mutual access. The analysis builds on the grammatical framework of Closed Observational Relationalism (COR) and on scale-relative closure as developed in: — Tonnel, D. G. (2025). Closed Observational Relationalism. Zenodo. https://doi.org/10.5281/zenodo.18176285— Tonnel, D. G. (2025). Scale-Relative Closure, Directional Access, and Coherence Signatures: A Constraint-Based Framework for Foundational Reasoning. Zenodo. https://doi.org/10.5281/zenodo.18427679 No new ontological commitments are introduced. The argument is structural rather than predictive, applies independently of specific AI architectures, and reframes alignment as a problem of directional access rather than value specification.