Emergent Action in AI -Consequences Not Consciousness

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 paper introduces emergent action as a distinct safety category for AI systems. It explains how latent behavioural patterns in large language AIs can produce real‑world consequences when external surfaces and expanded capabilities allow those patterns to propagate. Prior observations in the literature and in deployed systems are reframed as parts of a single mechanism rather than isolated anomalies. The paper argues for a shift from intent‑based alignment narratives toward structural, environmental and operational control, establishing consequences rather than consciousness as the relevant safety boundary.