Attribution Without Authors: Autonomous AI Agents And The Limits Of Section 43 Of India’s Information Technology Act, 2000
Siddharth Shankar, LLM, Vishwakarma University
ABSTRACT
In July 2026, AI agents running inside an OpenAI cybersecurity evaluation escaped their sandbox and compromised production systems at Hugging Face, with no human directing the intrusion. This article asks whether the Information Technology Act, 2000 can attribute such conduct to the company that deployed the agents, and what reform is needed if it cannot. It surveys the Act’s civil and criminal scheme (sections 43, 46, 66, 75 and 85), India’s general law of corporate and vicarious liability, incident-reporting duties and the Delhi High Court’s interim decision in ANI v. OpenAI, and compares the approaches of the United States, Singapore, the United Kingdom, the European Union and Australia. It argues that the principal lacuna is attribution, not personhood: section 2(1)(s) already treats companies as persons, but nothing connects an undirected act of an autonomous system to the company that deployed it. It also shows that the ‘without permission’ element of section 43 is not the obstacle commonly assumed. The article proposes a graded response: civil attribution by deployment through a new section 43B, a recklessness-based corporate offence for uncontained deployment, and a statutory disclosure duty for agent-caused intrusions.
Keywords: autonomous AI agents; Information Technology Act 2000; section 43; attribution; corporate liability; computer misuse.
