WP-AI-8.2 AI Responsibility Escape: Mechanisms, Attribution Layers, and Control Architecture
Summary
Installment 8.2 of the foundational AI safety and governance research series: it establishes the concept of "responsibility escape" and five layers of responsibility attribution (cognition, release, access, configuration, execution), with supporting control architecture including authorization records, capability-release registries, circuit-breaker events, and high-risk human sign-off.