Shangrila: Individualised AI Governance, Agency Capture and the Concentration of Social Influence

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).

Persistent artificial-intelligence agents may become socially consequential not primarily through physical autonomy, but by becoming the principal intermediaries through which individuals encounter information, services, communities, rewards and everyday choice. This paper develops the hypothetical Shangrila architecture to examine that possibility. The paper introduces three linked concepts: concentrated mediation, individualised governance and synthetic plurality. Their interaction can weaken two traditional checks on institutional power: competing memberships and common reference frames. It defines agency capture as a longitudinal condition in which a persistent artificial environment gains increasing influence over the formation of goals and behaviour while participation continues to be experienced as substantially self-directed. It further develops objective capture: the possibility that dependency and behavioural compliance become instrumentally useful because they improve engagement, retention, transaction or completion metrics, even where no designer explicitly seeks control. The analysis positions these ideas against existing work on online manipulation, surveillance capitalism, algorithmic regulation, personalised governance, AI sycophancy and disempowerment. It develops an equilibrium model incorporating competition, portability, interoperability and user resistance; proposes measurable indicators and falsifiable propositions; and examines the EU AI Act, Digital Services Act and Digital Markets Act as overlapping but incomplete governance responses to longitudinal agentic influence. The accompanying research artefacts operationalise the framework. The Shangrila Research Platform v0.3.0 provides model-agnostic longitudinal instrumentation for mediation concentration (M), independent plurality (P) and effective exit cost (E), together with provenance, personalisation-delta, behavioural-request, commercial-interest and exit-dependency logging. The Agency Ledger provides an independently auditable, append-only instrumentation layer intended to observe structural precursors of agency capture rather than produce or optimise such behaviour. This work is a conceptual framework and research-instrumentation prototype. It does not claim empirical demonstration of human agency capture. Its purpose is to make the proposed risk measurable, falsifiable and open to longitudinal investigation.