Fight, Flight, or Regulate? US Frontier AI, Regulatory Competition, and the Public Allocation of Risk

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

This conceptual preprint connects Extended Phenotype Theory (EPT), evolutionary game theory (EGT), and multilevel selection to US frontier AI governance and competition with China. It distinguishes enforceable cooperation, competitive protection and responsibility displacement as potentially coexisting mechanisms. A commitment game derives conditions under which compliance is an evolutionarily stable strategy without entry restrictions or liability insulation. A separate closure identifies conditions for sustaining oversight after compliance becomes prevalent. The analysis distinguishes legal decision rights from effective technical control of autonomous AI agents, and within-group selection from differential expansion between organizations, coalitions and jurisdictions. Stable compliance, private fitness and geopolitical success are not measures of public welfare. Historical comparisons, including Zenith and Argentine footwear, identify mechanisms without validating them in AI. Version 0.7. Formal deductions and synthetic consistency checks, not empirical calibration or validation. This preprint has not undergone independent peer review or specialist mathematical review. The author originated the research questions, central ideas and substantive arguments. The author is an attorney and Executive MBA, without formal mathematical training. AI provided substantial assistance with mathematical notation, equations, derivations and computational consistency checks, as detailed in the manuscript. The author retains responsibility for the manuscript. A footnote acknowledges Claudia Peirano for the conceptual contribution that prompted the expanded institutionalist analysis. Funding: none. Conflicts of interest: none declared by the author.