Abstract
As the trajectory of machine learning accelerates toward Artificial General Intelligence (AGI)—systems that equal or surpass human capabilities across all economically and strategically relevant domains—the international legal community faces an unprecedented regulatory imperative. This article provides a groundbreaking analysis of anticipatory jurisprudence, arguing that reactive, post-hoc legislative models are catastrophically inadequate for managing AGI due to the inherent potential for existential and systemic risk. The authors critically dissect the limitations of the European Union’s AI Act and the decentralized, sector-specific approaches adopted by the United States and the United Kingdom, demonstrating that these frameworks are designed for narrow AI and fundamentally fail to address the recursive self-improvement capabilities of AGI. By drawing jurisprudential parallels from the non-proliferation of nuclear weapons and the international regulation of recombinant DNA, the paper proposes a novel, supranational oversight architecture. We advocate for the establishment of an International AGI Regulatory Agency (IARA) endowed with unprecedented legal authority to conduct intrusive, on-site audits of hyperscale compute clusters and to enforce hard caps on algorithmic training runs utilizing massive computational thresholds. Furthermore, the article delves into the complex liability paradigms associated with autonomous superintelligence, questioning how traditional tort concepts of foreseeability, proximate cause, and strict liability can be applied when an AI system’s actions are deliberately opaque to its human creators. The authors conclude that averting global destabilization requires a paradigm shift from "innovation-first" deregulation to a precautionary legal regime that legally intertwines AGI development with provable alignment to fundamental human rights and international humanitarian law.
Keywords: Artificial General Intelligence, Existential Risk, AI Act, Anticipatory Jurisprudence, International Law, Machine Learning, Non-Proliferation