AI governance in India: the guidelines, the rules and who decides

AI governance in India runs on guidelines and existing law, not an AI statute. The India AI Governance Guidelines, released on November 5, 2025, say a separate AI law isn't needed at this stage. Existing IT, data protection, consumer protection, civil and criminal laws apply, and the 2026 IT Rules amendments cover synthetic content such as deepfakes.

Published and updated

Does India have an AI law?

No. The India AI Governance Guidelines, released on November 5, 2025, say a separate law to regulate AI isn't needed at this stage, given the current assessment of risks. They expect existing IT, data protection, consumer protection, civil and criminal laws to govern AI uses, through sector regulators, with targeted amendments where gaps appear.

Asked in the Lok Sabha on July 29, 2026 about a comprehensive AI law, the government pointed to the guidelines, the Information Technology Act, 2000 (IT Act) and the IT Rules (Unstarred Question 1645, opens in a new tab). It called the IT Act technology-neutral, so its provisions apply whatever the technology, AI included. Any new law or rule, if considered necessary, follows the government's established legislative procedures, it added.

The guidelines print no date of their own. A Press Information Bureau backgrounder, opens in a new tab dated February 15, 2026 describes them as "releasing in AI Impact Summit 2026", so you may see that date. The government gives the release date as November 5, 2025, in two Lok Sabha answers (Question 1645, opens in a new tab; Question 2988, opens in a new tab).

MeitY, the Ministry of Electronics and Information Technology, set up the committee that drafted the guidelines in July 2025 (executive summary, opens in a new tab).

What do the India AI Governance Guidelines say?

The India AI Governance Guidelines rest on seven principles and call for liability graded by function, risk and due diligence. They ask companies to follow Indian law, adopt voluntary codes, publish transparency reports and offer redress for AI harms. They advise regulators to avoid compliance-heavy rules, such as mandatory approvals or licensing, unless needed.

The seven principles, which the guidelines call sutras (Part 1, opens in a new tab):

  • Trust is the foundation
  • People first
  • Innovation over restraint
  • Fairness and equity
  • Accountability
  • Understandable by design
  • Safety, resilience and sustainability

On accountability, the guidelines favor a graded liability system based on the function performed, the level of risk and whether due diligence was observed (Part 2, opens in a new tab, section 2.5). They recommend clarifying how developers, deployers and end users are governed under existing laws such as the IT Act.

For anyone developing or deploying AI systems in India, Part 4, opens in a new tab recommends that you:

  • Comply with all Indian laws, including those on IT, data protection, copyright and consumer protection
  • Show that compliance when an agency or a sector regulator asks
  • Adopt voluntary principles, codes and standards
  • Set up a way to report AI-related harms, and resolve them within a reasonable time
  • Publish transparency reports on the risk of harm to people and society in India, shared with regulators in confidence if sensitive or proprietary
  • Explore techno-legal tools, such as privacy-enhancing technologies, machine unlearning and bias detection

For regulators, the same part advises against compliance-heavy requirements, such as mandatory approvals or licensing conditions, unless deemed necessary. It asks each regulator to pick the least burdensome tool that works, from industry codes to binding rules.

Do you need permission to launch an AI model?

No general rule read to October 31, 2026 requires government permission to launch an AI model. MeitY's advisory of March 15, 2024, which replaced one of March 1, 2024, asks platforms to label under-tested or unreliable AI models as possibly unreliable before offering them in India. The India AI Governance Guidelines advise regulators to avoid mandatory approvals unless needed.

The advisory, opens in a new tab (eNo.2(4)/2023-CyberLaws-3) says it supersedes MeitY's advisory of March 1, 2024. It asks every intermediary and platform to make under-tested or unreliable AI models available in India only after labeling the possible fallibility or unreliability of their output. A "consent popup" or a similar tool may tell users so (para 2(c)). It also asks that AI tools on a platform not let users share unlawful content, or permit bias, discrimination or threats to the integrity of the electoral process (paras 2(a) and 2(b)).

The guidelines add that more obligations may apply in specific contexts, such as sensitive applications or protecting vulnerable groups, and sector regulators keep their enforcement powers (executive summary, opens in a new tab, Part 2).

Existing duties still apply from launch. An intermediary whose service may let users create, change or share synthetic audio, images or video must use reasonable and appropriate technical measures, automated tools included, to stop unlawful synthetic content (IT Rules, opens in a new tab, rule 3(3)(a)(i)). That duty has applied since February 20, 2026 (G.S.R. 120(E), opens in a new tab).

Who steers AI governance in the government?

The AI Governance and Economic Group steers India's AI governance strategy. MeitY set it up on April 13, 2026, with the Minister of Electronics and IT as chairperson. Its terms include guidelines that hold firms to local laws, a study of regulatory gaps and the need for legal amendments, and sorting AI use cases into deploy, pilot and defer.

The Minister of State for Electronics and IT is vice chairperson, and MeitY's Secretary is member convener. The members include the Principal Scientific Advisor, the Chief Economic Advisor, the CEO of NITI Aayog and the Secretaries of Telecommunications, Economic Affairs, and Science and Technology (Office Memorandum No. ET/9/2023-ET-(Part 1), opens in a new tab, para (1)).

Its other terms of reference include:

  • Coordinating policy across ministries, departments and sector regulators
  • Working with industry on a roadmap for AI deployment over the next decade, its effect on jobs included

The government told the Lok Sabha on July 29, 2026 that, following the guidelines, it has also started two more bodies (Unstarred Question 1645, opens in a new tab). The Technology and Policy Expert Committee is to support technical and policy recommendations. The AI Safety Institute is to do safety research, develop standards, test AI systems and evaluate emerging risks.

Which existing rules reach AI products?

Three sets of rules reach AI products. Since February 20, 2026, the IT Rules regulate synthetic audio, images and video. The IT Act's intermediary rules apply now, depending on what your service does. From about May 13, 2027, the Digital Personal Data Protection (DPDP) Act governs the personal data your product processes, though not data the person made public.

On training, the guidelines add that copyright law may need amending, for example to allow large-scale training of AI models, while protecting copyright holders and the people whose data is used (overview of the framework, opens in a new tab). Copyright law itself wasn't among the documents read, so this page states no rule on it.

More in AI-generated content rules in India: labels, deepfakes, platforms and DPDP Act: what India's data protection law asks of a business.

Is there an AI certification in India?

Yes, for AI management systems. NABCB, the National Accreditation Board for Certification Bodies, accredits certification bodies that certify AI developers, producers and users to ISO/IEC 42001:2023, the government told the Lok Sabha on August 5, 2026. The answer doesn't say any law requires that certificate.

NABCB accredits those bodies to ISO/IEC 17021-1:2015 and ISO/IEC 42006:2025 (Lok Sabha Unstarred Question 2988, opens in a new tab).

How IndusDesk helps with AI governance in India

We watch the authorities, tell you what each change means for your products and plans, represent you while a decision is still open, and carry the work through, with every step on record in your workspace.

Frequently asked questions

Do the synthetic content rules cover AI-written text?

No: text on its own isn't synthetic information, though the general rules on unlawful content still cover it (MeitY's FAQ, opens in a new tab, Q8). Under rule 2(1)(wa) of the IT Rules, opens in a new tab, synthetically generated information is audio, images or video created or altered by computer so that it looks real and could pass for a real person or event. Routine editing, accessibility tools, translation, and good-faith documents, slides and training material don't count, as long as they create no false document and don't change the meaning (more in the AI-generated content guide).

Is your website, SaaS or AI chatbot an intermediary that can claim safe harbor?

It depends on what the service does: the IT Act, opens in a new tab calls an intermediary anyone who receives, stores or transmits a record for another person, or provides a service for that record (section 2(1)(w)). Whether an AI service qualifies, and can claim the section 79 safe harbor, depends on the nature of the service and its functions, the government told the Lok Sabha on July 29, 2026 (Question 1647, opens in a new tab). Among the documents read to October 31, 2026, it issued no list; the IT Rules guide covers the due diligence the safe harbor needs.

Do the synthetic content rules cover an enterprise AI tool?

Only if the tool is an intermediary whose service may let users create, change or share synthetic content (IT Rules, opens in a new tab, rule 3(3)). Whether an enterprise AI tool is an intermediary at all depends on the nature of the service, the government says; among the documents read to October 31, 2026, it issued no list. A covered service must also warn users that misuse can bring penalties under the IT Act, the Bharatiya Nyaya Sanhita, the POCSO Act and other laws (rule 3(1)(ca)).

Sources