India AI Governance Guidelines
Applies to: AI developers, deployers and regulators across sectors (non-binding framework)
India's national framework for AI governance, developed by a MeitY drafting committee constituted in July 2025 (chaired by Prof. Balaraman Ravindran) and published on 15 February 2026 ahead of the India AI Impact Summit, building on an initial unveiling in November 2025. It is explicitly non-binding — a principle-based, "techno-legal", light-touch framework that leans on existing laws rather than new AI legislation. It is anchored in seven "sutras": Trust is the Foundation, People First, Innovation over Restraint, Fairness & Equity, Accountability, Understandable by Design, and Safety-Resilience-Sustainability — adapted from the RBI's August 2025 FREE-AI Committee report. "Innovation over Restraint" is the deliberate philosophical stake, positioned as a counterpoint to the EU AI Act's precautionary, risk-classification-first model. The guidelines recommend three new institutions — an AI Governance Group (AIGG) for inter-agency coordination, a Technology & Policy Expert Committee (TPEC) for technical/legal advice, and an IndiaAI Safety Institute (AISI) for model testing and standards — and reference CERT-In, I4C and NCIIPC in the risk-mitigation pillar. A proposed AI incident database would record misuse, harmful outputs and security failures.
Why it matters
Read this as direction-of-travel, not obligation — nothing here is enforceable today, and that is the point. The signal for a CII CISO is the "innovation over restraint" stance: India is deliberately not building an EU-style risk-tiered AI law, so near-term AI accountability will be enforced through the instruments you already track — DPDP, the CERT-In AI Blueprint, the IT Amendment Rules on synthetic media, and sectoral advisories — rather than a single AI Act. The detail worth noting is lineage: the seven sutras come from the RBI's FREE-AI report, so financial-sector AI thinking is quietly setting the national tone, and the AISI's testing standards are the thing to watch — once it publishes model-evaluation expectations, they will likely flow into RBI/SEBI AI advisories as concrete controls. Treat this as the map of where binding rules will harden next.