Instructions, Consequences, and Human Judgment: Normativity and Liability in Agentic Systems through Extended Phenotypes and Evolutionary Games

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 paper examines the relationship between instructions to AI agents, institutional powers, consequences, execution controls, and civil liability. Starting from Kelsen, it engages Hart, Carrió, Dworkin, Raz, Fuller, Schauer, and Ostrom without treating their disagreements as a single doctrine. It develops the author's extended-phenotype and evolutionary-game-theoretic programme through a candidate normative-operational package linking instructions, competences, evidence, review, and consequences. Legal validity, justification, behavioural effectiveness, containment, and differential retention remain distinct. A candidate architecture preserves authorization conditions until actions produce effects. Documented incidents and prior experiments motivate hypotheses rather than validate the architecture. The paper offers discriminating predictions, rival explanations, and separate technical, behavioural, human, and evolutionary evaluation requirements. It reports no new experiments and has not undergone independent peer review.