The Limits of Artificial Agency: Structural Conditions for Artificial Identity and Purpose

Discussions of artificial intelligence routinely treat goal pursuit, agency, intention, and purpose as equivalent descriptions of increasingly capable systems. This paper argues that these concepts are not equivalent, and that treating them as such has important consequences for how AI systems are interpreted and aligned. Drawing on the Coherent Arc Model of Purpose, it provides a structural account of what purpose consists of and locates current artificial systems precisely within that account. Purpose, on this analysis, is the regulation of an abstract, evaluative, temporally extended self-identity toward coherent completion within constraint. It arises only when three architectural conditions are jointly satisfied: persistent self-representation, evaluative identity structure, and arc coherence regulation operating as a motivationally operative system. Current AI systems, including large language models and reinforcement learning architectures, do not satisfy these conditions. Behaviours frequently interpreted as evidence of emerging agency, including metric gaming, strategy preservation, and apparent self-interest in continued operation, are better understood as trajectory stabilisation within optimisation systems, a pattern that arises across evolutionary, mechanical, and computational substrates without implying identity-arc regulation. The paper identifies the structural threshold at which artificial purpose would become possible, specifies what would be architecturally entailed if that threshold were crossed, and draws out the implications for alignment research. The central claim is that optimisation and purpose are structurally distinct regulatory processes, and that this distinction does not collapse as optimisation becomes more sophisticated. Identifying the transition between regimes, if it occurs, requires attending to architectural features rather than behavioural ones.