Duggal Global Agentic AI Liability Framework: A Conceptual, Normative, and Operationally Precise Blueprint for Accountability, Responsibility, and Governance in the Era of Autonomous Artificial Intelligence Agents (Version 1.1)

This Framework addresses the global liability vacuum created by the transition from Generative AI to Agentic AI, being systems that autonomously set goals, plan, select and execute tools, retain memory, coordinate across agents, and adapt in real time without continuous human direction. It argues that existing legal doctrines including tort, product liability, contract, criminal law, and current AI-specific regulation, were built for human actors and static tools, and fail precisely at the point where Agentic AI operates: emergent, multi-step, tool-executing autonomy across jurisdictions. The Framework proposes a five-tier liability stack (strict liability, presumptive/Black Box Barrier liability, negligence under a Reasonable Agentic AI Governance Standard, joint-and-several pipeline liability, and regulatory/administrative liability), a five-level autonomy taxonomy tying legal duty to degree of autonomy granted, a role-based lifecycle allocation matrix spanning Developer, Provider, Integrator, Deployer, Operator, and User, and a five-tier harm classification system (Prohibited through Low Risk). Its central jurisprudential contribution is a set of fifteen original doctrines addressing discrete liability questions- including hallucination, fine-tuning, retrieval-augmented generation, memory persistence, tool use, model drift, multi-agent delegation, and cross-agent amplification- that existing AI law has not yet systematically resolved. The Framework is offered as a model for national legislation, a template for international instruments, an enterprise governance standard, a judicial reference tool, and an insurance underwriting standard. It includes a comparative jurisdictional analysis across the EU, US, UK, India, Singapore, UAE, Australia, Canada, and China. Version 1.1 incorporates calibrated language revisions to Version 1.0 (March 2026), affecting the Preamble, Executive Summary, Sections 1.2 and 4.3, and the Conclusion. The substantive architecture is unchanged. This document is a conceptual and normative policy framework, published for stakeholder consultation, scholarly reference, and policy development purposes. It does not constitute legal advice.