ORCHESTRATING ACCOUNTABILITY: LIABILITY OF THE SYSTEM DEPLOYER IN AGENTIC AI UNDER INDIAN LAW
ORCHESTRATING ACCOUNTABILITY: LIABILITY OF THE SYSTEM DEPLOYER IN AGENTIC AI UNDER INDIAN LAW Samarth Udasin, 5th year student of B.Sc.LL.B (Hons.) [Cyber Security] at National Law Institute University, Bhopal (India). Divyansh Rai, 5th year student of B.Sc.LL.B (Hons.) [Cyber Security] at National Law Institute University, Bhopal (India). Priyanshu Tripathi, 5th year student of B.A. LL.B (Hons.) at National Law Institute University, Bhopal (India). Download Manuscript doi.org/10.70183/lijdlr.2026.v04.306 Agentic AI systems do not merely respond to queries they act, decompose goals, call APIs, enter transactions, and produce irreversible real-world consequences autonomously. That shift from tool to actor breaks every assumption on which India’s intermediary liability law rests. This paper argues that existing Indian law already assigns liability to the right party: the System Deployer, the entity that selects the AI model, integrates tools, engineers data flows, and directs the agent’s consequential actions. The principle is straightforward liability follows control. Section 79 of the Information Technology Act, 2000 shows that the Deployer’s active orchestration role disqualifies it from safe harbour: while it qualifies as an intermediary under the statutory definition, it fails every substantive condition that Section 79(2) requires before that status translates into immunity. The Digital Personal Data Protection Act, 2023 confirms this, designating the Deployer a Data Fiduciary who will carry non-delegable duties once Section 8 commences. SEBI’s algorithmic-trading framework demonstrates that Indian regulators already impose full liability on deploying entities regardless of algorithmic opacity. Tort principles of negligence and absolute liability complete the picture. Comparative analysis of the EU Digital Services Act and L’Oréal v eBay confirms global convergence. The paper concludes with a governance framework grounded in the RBI FREE-AI Report. No new legislation is needed.