The Data Security Council of India (DSCI) has introduced a detailed industry framework aiming to clarify liability when autonomous AI agents cause harm, emphasizing distributed responsibility across multiple participants rather than focusing solely on deployers or providers.
- Liability linked to control over AI decisions, from deployment to oversight
- Framework uses existing laws rather than creating new AI-specific regulations
- Six safeguards recommended, including permissions restriction and human approval
What happened
The Data Security Council of India (DSCI) published an industry framework titled ‘AI Liability: Industry Guidance for Agentic AI’ that proposes assigning liability for harm caused by autonomous AI agents across the entire AI value chain. This includes various involved groups such as end-user organisations, service providers, product companies, tool providers, and infrastructure hosts. The framework seeks to clarify legal responsibility by analyzing who controls decisions like model selection, configuration, and human oversight.
Unlike traditional AI, agentic AI systems perform tasks autonomously—interacting with APIs, making decisions, and delegating actions with minimal human review. This evolution raises new risks such as unauthorized transactions, data breaches, and cascading errors, especially when multiple AI agents interact. DSCI highlights 12 potential harm areas including privacy violations, discrimination, and AI-enabled cybercrime, underscoring the need for clear accountability mechanisms.
Why it matters
DSCI’s framework aims to address a significant gap in AI governance by proposing liability allocations based on participants’ control rather than defaulting responsibility solely to AI deployers or technology providers. This approach aligns liability with practical influence, addressing complex scenarios where multiple entities shape AI behavior and outcomes. It also emphasizes leveraging existing laws like duty of care and product liability to reduce regulatory uncertainty and fragmentation.
The rise of agentic AI systems, which operate with greater autonomy than traditional generative AI, poses unique operational challenges, such as model behavior changes and delegation between agents. These factors complicate liability assessments but are crucial to mitigate risks associated with autonomous technologies. Establishing clear liability frameworks influences how companies manage AI safety, transparency, and contract terms, fostering more responsible AI adoption in the Indian market.
What to watch next
Additionally, the framework distinguishes liability depending on the AI model type and deployment context. Companies running open-source models on their infrastructure bear heavier auditing responsibilities, while those using closed API-based models primarily rely on contractual safeguards. Future regulatory or judicial responses may further clarify these assignments, influencing AI providers, integrators, and users across India’s growing AI ecosystem.