Abstract This paper argues that the lawful deployment of agentic AI systems requires demonstrated technical and operational capacity to respect jurisdictional legal boundaries. This is a capacity requirement, not a moral claim. Drawing on binding legal frameworks in Australia, the European Union, and international law, the paper maps existing enforceable duties onto the operational requirements for agentic AI systems. The analysis covers child protection, disability rights, domestic violence and coercive control, privacy, Indigenous data sovereignty, and environmental protection. The paper concludes that flooding markets with AI systems that lack containment capacity is incompatible with existing law, and that kill switches represent standard safety infrastructure analogous to recalls, grounding orders, and emergency stops in other regulated industries.
John R. Smith (Fri,) studied this question.