Who Is Liable When AI Commits a Crime? Mens Rea in the Age of Autonomous Systems

Who Is Liable When AI Commits a Crime? Mens Rea in the Age of Autonomous Systems

The Problem of the Absent Mind

Criminal law rests on an ancient maxim: actus non facit reum nisi mens sit rea, an act does not make a person guilty unless the mind is also guilty. For centuries this principle presupposed a human actor whose intention, knowledge or recklessness could be examined. Autonomous systems unsettle that premise. When a trading algorithm manipulates a market, a chatbot generates defamatory or inciting content, or an AI agent executes a fraudulent transaction that no human specifically instructed, the law confronts an uncomfortable question: whose mind was guilty?

Why the Machine Cannot Be the Accused

Under the Bharatiya Nyaya Sanhita, 2023, a “person” includes companies, associations and bodies of persons, but not software. An AI system possesses neither legal personality nor the capacity for intention in the sense criminal law demands; it cannot be summoned, punished or reformed. Prosecuting the machine would therefore be both conceptually incoherent and practically futile. Responsibility must ultimately be traced to the human and corporate actors who design, train, deploy and profit from it.

Three Models of Attribution

Comparative scholarship, notably the work of Gabriel Hallevy, offers a useful taxonomy. Under the first model, the AI is an innocent agent, and the person who uses it to commit an offence is the real perpetrator, much as one who employs an unknowing intermediary. Under the second, liability rests on natural and probable consequences: a developer or deployer who could reasonably have foreseen the harmful output may be held responsible for negligence or recklessness. The third model contemplates direct liability of the system itself, a proposition that remains largely theoretical and has found no footing in Indian law.

Indian law already contains tools that accommodate the first two models. Corporate criminal liability is well established: Standard Chartered Bank v. Directorate of Enforcement (2005) held that a company may be prosecuted even for offences carrying mandatory imprisonment, and Iridium India Telecom v. Motorola (2011) accepted that the intent of those who control a company may be attributed to it. Section 85 of the Information Technology Act, 2000 renders persons in charge of a company liable for offences committed by it, unless they prove lack of knowledge or due diligence. Section 79 conditions an intermediary’s immunity on due diligence, a standard now sharpened by the 2026 amendments on synthetically generated information.

The Remaining Gap

What the law lacks is a coherent standard for foreseeability in systems whose behaviour emerges from training rather than explicit instruction. The India AI Governance Guidelines released by the Ministry of Electronics and Information Technology in November 2025 deliberately favour existing laws over a new prescriptive statute, and emphasise accountability and safety among their guiding principles. That is a reasonable starting point, but principles must eventually harden into enforceable duties: documented risk assessments, audit trails, human oversight for high risk uses, and clear allocation of responsibility across the AI supply chain.

Conclusion

The absence of a guilty mind in the machine must never become an absence of accountability in society. Criminal law has adapted before, to corporations, to strict liability and to cyberspace itself. It must now adapt again, ensuring that those who unleash autonomous systems remain answerable for the foreseeable harms they cause.