DSCI Proposes Liability Framework for Autonomous AI Agents

The Data Security Council of India (DSCI) has published a discussion paper, 'AI Liability: Industry Guidance for Agentic AI,' proposing a structured approach to assigning responsibility when autonomous AI agents cause damage. Rather than introducing new legislation, the framework suggests applying existing legal principles—such as duty of care and product liability—to define accountability across the AI value chain.

The document categorizes participants into five groups, including end-user organizations, service integrators, and infrastructure providers. The DSCI argues that liability should be determined by control over specific decisions, such as system configuration, tool permissions, and human oversight. This shift is critical as AI transitions from passive generative models to autonomous agents capable of executing tasks and interacting with APIs with minimal human intervention.

Potential risks identified by the council include data breaches, financial errors, and cybercrime. To address these, the framework proposes layered safeguards, including the implementation of 'kill switches' and mandatory human approval for critical actions. The proposal is currently open for industry feedback, reflecting broader efforts by Indian regulators to establish governance for emerging technologies.