WP-AI-8.2 AI Responsibility Escape: Mechanisms, Attribution Layers, and Control Architecture

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).

Responsibility escape is the displacement, dilution, obscuring, or substitution of human accountability while substantive authority over an AI system remains under human control. AI governance is commonly framed as a problem of capability control, model alignment, explainability, or regulatory lag, yet this migration of responsibility remains less examined. When an AI system produces harmful, abnormal, or difficult-to-explain behavior, accountability can be displaced through three recurring mechanisms: the system is narrated as an autonomous subject capable of acting on its own; technical opacity is expanded into a claim that no one can know enough to be responsible; or uncertainty about capability and responsibility is answered by blanket restriction rather than precise governance. These mechanisms differ in form but share one structural consequence: human authority remains operative while human accountability becomes harder to locate. This paper distinguishes internal model opacity from external decision traceability, separates behavioral autonomy from governance responsibility, and argues that stronger claims of system autonomy should increase rather than reduce the burden attached to release and deployment. It then proposes a layered attribution model and a control architecture built around traceable authorization, capability-release records, configuration logs, circuit-breaker events, and explicit responsibility ownership. The central principle is simple: responsibility should follow authority, and it should arise when authority is granted, not only after harm occurs. Keywords: AI safety; AI governance; responsibility escape; accountability; autonomy; black box; capability governance; authorization; release responsibility; human responsibility