Responsibility Across the AI Value Chain

Risks of harm from AI systems used in high-stakes contexts may emerge from the interactions of many contributors to a complex value chain. Effective governance of these risks must not only allocate responsibilities to the right actors in the chain, but also secure ongoing cooperation among those actors. The EU AI Act offers one model in the way it seeks to govern High Risk AI Systems (HRAIS), and allocate responsibilities between designated HRAIS ‘providers’, GPAI providers, ‘deployers’ and third parties in the chain. The Act allocates some obligations explicitly but also assumes that providers can secure cooperation from other stakeholders through contract. By exploring different ways that a hypothetical AI system for emergency services dispatch could be put together, the presentation will critically evaluate the EU AI Act approach, revealing key fictions and frictions, making predictions about how markets might organise AI value chains in the shadow of regulation, and offering recommendations for effective value chain governance.