Criminal Liability for Autonomous AI-Generated Crimes: Accountability Challenges in Pakistan's Criminal Justice Syste

Abstract

The proliferation of autonomous artificial intelligence (AI) systems ranging from self-driving vehicles and algorithmic trading bots to autonomous weapons and generative AI models has introduced a category of harmful conduct that does not fit comfortably within the doctrinal architecture of criminal law. Criminal liability has traditionally rested on two pillars: a wrongful act (actus reus) and a culpable mental state (mens rea), both of which presuppose a human or corporate agent capable of intention, knowledge, or recklessness. Autonomous AI systems complicate this picture because their outputs may not be foreseeable even to their designers, and the causal chain linking human conduct to the eventual harm may be attenuated by layers of machine learning, self-modification, and emergent behaviour. This article examines how Pakistan's criminal justice system anchored in the Pakistan Penal Code 1860, the Prevention of Electronic Crimes Act 2016, and general principles of criminal responsibility inherited from the common law is positioned to respond to AI-generated crimes. It surveys competing theories of AI liability developed in comparative scholarship (respondeat superior analogies, natural-probable-consequence liability, direct liability, and proposals for a novel legal status for AI), and applies them to Pakistan's statutory and institutional context. The article argues that Pakistan's existing framework is structurally unable to attribute mens rea to autonomous systems, that corporate and vicarious liability doctrines offer only partial solutions, and that investigative and forensic capacity within the police and judiciary lags behind the technical complexity of AI systems. It concludes with a set of legislative, institutional, and procedural reforms aimed at closing the accountability gap without stifling innovation.