AI agent liability should be shared across the AI value chain according to who controls each decision, the Data Security Council of India (DSCI) proposes in a new industry framework.
AI agent liability: what DSCI proposes
- The discussion paper, ‘AI Liability: Industry Guidance for Agentic AI’, applies existing laws, contracts and sectoral rules instead of creating a new regulation.
- DSCI says blaming only the company that deploys an AI system, or only the technology supplier, is not enough. Liability should follow control over model choice, configuration, tool access, permissions and human oversight.
- An ‘authority map’ would show who controlled a decision and whether they could have prevented the harm.
Who is responsible for what
- Deploying companies: use cases, permissions, risk checks and when a human must approve.
- Product companies: product claims, default settings and model versions. Integrators: system design, approval steps and failure handling.
- Tool and framework providers: orchestration, memory and safety features. Cloud providers: isolation, identity, data retention and logs.
- Model providers: capabilities, safety measures, disclosure of known limits and notice of major behaviour changes, with less liability when others take control of deployment.
Self-hosted models carry more responsibility
Firms running open-weight models on their own infrastructure take on more duty to audit, modify and monitor them. Those using closed models through managed APIs may seek recourse from the provider, depending on contract terms and liability caps.
Six safeguards DSCI recommends
- Remove capabilities the agent does not need and restrict its permissions.
- Require human approval for significant or irreversible actions.
- Monitor behaviour, let systems scale back safely and assign clear owners to kill switches.
- Assess risks before deployment, add liability clauses to procurement contracts and review safeguards regularly.
What this means for you
This is voluntary guidance, not law. If your business lets AI agents handle payments, data or customer requests, review their permissions and your vendor contracts now. DSCI has invited feedback; a deadline was not stated.
AI agent liability FAQ
What is the DSCI AI agent liability framework?
It is a discussion paper that proposes assigning responsibility for AI agent harm across the value chain, based on each party’s control over key decisions.
Is the DSCI AI agent liability framework a law?
No. It is industry guidance that applies existing laws and contracts, and DSCI has invited feedback from developers, industry and policy groups.
Who is liable when an AI agent causes harm under the framework?
DSCI says no single party is automatically liable. Responsibility depends on who controlled the relevant decision and could have limited the harm.
Sources
- Inc42 – DSCI Floats Framework To Determine Liability For Harm Caused By AI Agents – 4 Nov 2026
- Economic Times – Accountability tames AI autonomy as DSCI outlines its liability framework – date not shown in the excerpt read
Related: Goa Launches AI Policy 2026, Targets 50,000 Goans Trained in AI by 2030
Last updated: 5 November 2026
