India does not have a separate AI law, and the government has decided it does not need one yet. The India AI Governance Guidelines, released by MeitY on 15 February 2026 during the AI Impact Summit in New Delhi, set out seven principles and a set of new institutions. AI is otherwise regulated through laws that already exist, such as the IT Act, the DPDP Act and sector rules from RBI, SEBI and IRDAI.

For working professionals, that has a practical effect. No single “AI licence” exists. Instead, the obligations come from data protection, content rules and your sector’s regulator. Here is what the guidelines say, what they do not say, and what they mean for your job.

Why India chose guidelines, not a new law

The European Union passed a dedicated AI Act that sorts AI systems into risk categories with fixed obligations. India’s committee, set up by MeitY in July 2025, took a different view. Its position is that most AI harms, such as fraud, discrimination, privacy breaches and deepfakes, are already covered by existing law. Where gaps appear, the guidelines recommend targeted amendments rather than a broad new statute.

The government describes this as a principle-based approach that favours innovation, backed by a “techno-legal” toolkit: standards, audits, sandboxes and technical safeguards built into systems. A new law is not ruled out. The long-term action plan explicitly mentions adopting new laws for emerging risks if they are needed.

The seven principles (“sutras”)

PrincipleWhat it means in practice
Trust is the foundationTrust has to run across the AI value chain: data, models, deployment and use
People firstHumans keep meaningful oversight, especially where decisions affect people
Innovation over restraintResponsible innovation is preferred to precautionary bans
Fairness and equityOutcomes should not discriminate, with attention to marginalised groups
AccountabilityDevelopers and deployers are answerable for how their systems behave
Understandable by designDisclosures and explanations should be built in, not added later
Safety, resilience and sustainabilitySystems should be secure, robust and environmentally responsible

These are principles, not legal duties. They matter because regulators and courts can use them as a reference point when they interpret existing rules for AI cases.

Which laws already apply to AI

  • IT Act, 2000 and the IT Rules. The February 2026 amendments require labelling of synthetically generated audio, images and video, and shorten takedown deadlines. See our explainer on India’s 2026 deepfake rules.
  • Digital Personal Data Protection Act, 2023 and the 2025 Rules. Any AI system trained on, or making decisions about, personal data has to follow consent, purpose-limitation and security duties. Our DPDP Act explainer covers what changes at work.
  • Sector regulators. RBI, SEBI and IRDAI oversee AI used in lending, investment advice and insurance through their existing rules on outsourcing, fair treatment and model risk.
  • General law. Consumer protection, copyright, contract and criminal law still apply when AI causes loss.

New institutions

The guidelines recommend a light, coordinating structure rather than a new regulator:

  • AI Governance Group (AIGG): a whole-of-government body to coordinate policy across ministries and regulators.
  • Technology & Policy Expert Committee (TPEC): outside experts who advise the AIGG.
  • IndiaAI Safety Institute: research, testing, standards and technical guidance. The guidelines say it needs adequate resources to do this.
  • MeitY remains the nodal ministry. NITI Aayog continues to anchor the national AI strategy.

The action plan: what happens when

The plan is phased, but it gives no calendar dates for any phase. Expect these steps to arrive gradually.

PhaseMain steps
Short termSet up AIGG and TPEC, build an India-specific risk framework, run gap analyses of existing laws, issue a master circular, start awareness programmes
Medium termPublish common standards, run a national AI incident database, amend laws where needed, pilot regulatory sandboxes in high-risk areas
Long termMonitor and review the framework, adopt new laws for emerging risks, expand international engagement

IndiaAI Mission: compute and data

Governance sits alongside public investment. The Union Cabinet approved the IndiaAI Mission in March 2024 with an outlay of ₹10,371.92 crore over five years. At the time of the guidelines, the government said over 38,000 GPUs had been onboarded, against a target of 100,000. They are offered to startups, researchers and students at subsidised rates through a common compute portal.

The AIKosh platform lists more than 9,500 datasets and 273 sectoral models. The government also reported support for 500 PhD scholars, 5,000 postgraduates and 8,000 undergraduates, plus 570 AI data labs. If you work in a startup or a research team, the compute portal and AIKosh are worth checking before you pay commercial cloud prices for experiments.

What it means for companies

A simple test helps. Take a bank deploying an AI model that screens loan applications. No AI licence is needed. But the bank still has to:

  1. Process applicants’ personal data lawfully under the DPDP Act, with notice and a valid purpose.
  2. Follow RBI’s rules on fair lending, outsourcing and grievance redressal, because AI does not change the bank’s accountability.
  3. Be able to explain a rejection to the customer in plain terms (the “understandable by design” principle).
  4. Keep a person in the loop for edge cases and complaints (“people first”).
  5. Log incidents and fix bias it finds. A national AI incident mechanism is planned.

Companies that already run a responsible-AI programme will find little new here. Those that have deployed chatbots or models without documentation should start with an inventory: which AI systems are live, what data they use, who owns them and what could go wrong. Agentic AI, where software takes actions rather than just answering, needs extra controls. Our guide to securing AI agents at work lists them.

What it means for your career

  • AI governance, risk and compliance (GRC): companies need people who can map AI systems to DPDP, IT Rules and sector rules. Lawyers, auditors and compliance staff with some technical fluency are well placed.
  • Model evaluation and red-teaming: testing for bias, safety and misuse is becoming a standard step before launch.
  • Standards: ISO/IEC 42001, the international standard for AI management systems, gives companies a ready-made structure. Familiarity with it, and with the NIST AI Risk Management Framework, is useful on a CV.
  • Data protection: DPDP compliance work overlaps heavily with AI governance.

Common misunderstandings

  • “No AI law means no rules.” Wrong. Existing laws apply in full, and penalties under the DPDP Act alone can reach ₹250 crore per breach.
  • “The guidelines are binding.” The seven principles are not legal obligations by themselves. The IT Rules amendments on synthetic content, by contrast, are binding.
  • “This copies the EU AI Act.” It does not. There is no risk-tier classification with fixed duties, though the guidelines do call for graded obligations based on risk.

FAQ

Do I need government approval to launch an AI product in India?

Not under the guidelines. You do need to comply with sector rules. For example, a lending app needs RBI-compliant arrangements whether or not it uses AI.

Will India pass an AI Act later?

Possibly. The plan keeps that option open for risks existing law cannot handle, but no draft bill has been announced.

This article is general information, not legal advice. For decisions about a specific AI deployment, consult a qualified lawyer or compliance professional.

Last reviewed: 8 October 2026.

Sources

Leave a Reply

Trending