Skip to content
MeitY Released India's AI Governance Guidelines on 5 November 2025: Seven Sutras, Six Pillars and a Plan That Builds on Laws India Already Has
AI & Technology

MeitY Released India's AI Governance Guidelines on 5 November 2025: Seven Sutras, Six Pillars and a Plan That Builds on Laws India Already Has

Illustration by tuput

English

The guidelines, drafted by a committee chaired by IIT Madras Professor Balaraman Ravindran, say a separate AI law is not needed at this stage and lean on existing law, voluntary codes and technical safeguards. A new governance group, an AI incident database and regulatory sandboxes are among the steps they propose.

· · 9 min read

On 5 November 2025 the Ministry of Electronics and Information Technology (MeitY) released the India AI Governance Guidelines under the IndiaAI Mission. Principal Scientific Adviser Prof. Ajay Kumar Sood unveiled them, according to the Press Information Bureau (PIB) release. The framework has four parts: seven guiding principles called sutras, recommendations across six pillars, an action plan with short, medium and long-term steps, and practical guidance for industry and regulators.

Its central choice is a plan built on laws India already has. The document states that “at this stage, a separate law to regulate AI is not needed given the current assessment of risks.” MeitY Secretary S. Krishnan said at the launch, “Our focus remains on using existing legislation wherever possible.” The guidelines are recommendations, not law, and every duty described below is a recommendation.

Who wrote it and how long it took

The groundwork began with a high-level advisory group on AI set up in 2023 under the Principal Scientific Adviser. Its sub-committee on AI governance, which included Prof. Balaraman Ravindran, Debjani Ghosh and Sharad Sharma, prepared a draft report that MeitY published for comment. According to the guidelines’ own annexure, more than 2,500 submissions came in from government bodies, universities, think tanks, industry associations, companies and individuals.

MeitY then formed a drafting committee in July 2025. Ravindran, a professor at IIT Madras, chaired it. Its members included Additional Secretary Abhishek Singh (also CEO of the IndiaAI Mission), Debjani Ghosh of NITI Aayog, Rahul Matthan, a partner at Trilegal, and Sharad Sharma, co-founder of the iSPIRT Foundation. Amlan Mohanty of NITI Aayog was the lead writer. At the launch, Singh said a second committee reviewed the public input to refine the final text.

Seven sutras, adapted from a banking report

A sutra here is a short guiding rule. The seven are Trust is the Foundation, People First, Innovation over Restraint, Fairness and Equity, Accountability, Understandable by Design, and Safety, Resilience and Sustainability. The guidelines say these were adapted from the principles of the Reserve Bank of India’s FREE-AI Committee, a panel the RBI set up on 26 December 2024 to frame the use of AI in finance. The RBI’s press release of 13 August 2025 says that committee developed seven sutras and 26 recommendations under six pillars.

The third sutra states the priority: “All other things being equal, responsible innovation should be prioritised over cautionary restraint.” The second, People First, says humans should “as far as possible” have final control over AI systems. Prof. Sood described the spirit of the whole framework in two words in the PIB release: “Do No Harm.”

Six pillars and what each asks for

The recommendations sit under six pillars, grouped into three domains: enablement (infrastructure and capacity building), regulation (policy and regulation, and risk mitigation) and oversight (accountability and institutions).

  • Infrastructure: wider access to data, computing power and digital public infrastructure (DPI, the shared systems such as Aadhaar and UPI that Indian services build on). The document reports that, as of 31 August 2025, over 38,231 GPUs were available to startups and researchers at subsidised rates and that the AIKosh platform held 1,500 datasets and 217 models. A separate tuput article covers the IndiaAI Mission’s compute, models and money.
  • Capacity building: training for citizens, officials, regulators and police units that investigate AI-enabled crime, with more reach into tier-2 and tier-3 cities.
  • Policy and regulation: review existing laws, find gaps and fix them with targeted amendments.
  • Risk mitigation: an India-specific risk framework, an AI incident database, voluntary measures and technical safeguards.
  • Accountability: graded liability based on the function a party performs, the level of risk and whether it exercised due diligence, plus grievance channels.
  • Institutions: a whole-of-government approach, with sector regulators keeping their enforcement powers.

Why the guidelines say existing law is enough for now

The committee reviewed laws on information technology, data protection, intellectual property, competition, media, employment, consumer protection and crime. It concluded that many AI risks can be handled under them. Its examples: deepfake impersonation falls under the Information Technology Act and the Bharatiya Nyaya Sanhita, and using personal data without consent to train a model is governed by the Digital Personal Data Protection Act.

The same chapter lists where the committee sees gaps or open questions:

  • Classification and liability. Section 79 of the IT Act gives platforms immunity for unlawful content posted by third parties, provided they do not initiate, select or modify the transmission. The guidelines say that immunity “would not be applicable to many types of AI systems that generate or modify content,” and recommend amending the Act to say how AI systems are classified and who is liable.
  • Data protection. The guidelines raise questions on training with publicly available personal data and on how purpose limits fit modern AI, and suggest the new governance group review them. At the time of writing, the document said the DPDP Act would come into force once draft rules were notified.
  • Content authentication. The guidelines call harmful deepfakes a growing menace, point to the C2PA industry standard for content provenance, and recommend an expert committee to develop standards.
  • Copyright. The guidelines note that the fair-dealing exceptions in Section 52 of the Copyright Act are limited to non-commercial use and may not cover much AI training. They suggest the Department for Promotion of Industry and Internal Trade (DPIIT) committee, set up in April 2025, consider a text and data mining exception, meaning permission to copy works in bulk to train models, with protections for rights holders.

Courts are already hearing copyright disputes over training data, and tuput has covered three rulings from 2026.

The guidelines favour voluntary measures: industry codes, technical standards and self-certification. They say some baseline measures may later become mandatory, enforced by sector regulators. Industry is asked to follow all Indian laws, publish transparency reports, set up grievance redressal and explore technical safeguards.

“Techno-legal” is the guidelines’ term for building a legal requirement into the design of a system, so that “digital architecture enforces what law requires.” The document’s worked example is DEPA, the Data Empowerment and Protection Architecture, which shares data through consent tokens and was first used in finance. The guidelines suggest adapting it to AI training, and they name the trade-offs: privacy-protecting methods can lower a model’s score on some benchmarks, and the approach does little to govern a model’s effects after training.

The incident database is the most concrete new tool. The guidelines propose a national, federated system, meaning one central database that can query smaller sector databases built to a common format. They suggest reporting without the threat of penalties, so that policymakers can see what kinds of harm AI causes and how. The document uses the OECD’s definition of an AI incident, an event where the development, use or malfunction of an AI system directly or indirectly leads to a specific harm. It also proposes regulatory sandboxes, controlled test environments that carry reasonable legal immunities if the tests produce published evidence of what was tested and what risks appeared.

On risk, the guidelines list six categories: malicious use, bias and discrimination, transparency failures, systemic risks, loss of control and national security. They single out children, and women, who the document says bear the brunt of AI-generated deepfakes.

Institutions and the action plan

The guidelines propose an AI Governance Group, a small permanent inter-agency body, supported by a Technology and Policy Expert Committee. The AI Safety Institute under the IndiaAI Mission would test and evaluate AI systems and advise on safety. The guidelines give an illustrative membership for the group, chaired by the Principal Scientific Adviser, with MeitY, the Home and External Affairs ministries, the Department of Telecommunications and regulators such as the RBI, SEBI and TRAI.

The action plan puts the groups, the risk framework, a gap analysis of laws and a master circular on applicable regulations in the short term. Common standards, the national incident database, law amendments and sandbox pilots come in the medium term. Reviews and, if emerging risks require, new laws sit in the long term.

What has followed

Three later developments connect to the guidelines, each reported by the sources below.

  • In April 2026, according to reports from that month, MeitY constituted the AI Governance and Economic Group. Construction World (22 April) reports it as the apex inter-ministerial body, chaired by the Union Minister for Electronics and IT, with the Minister of State as vice-chairperson, and says its constitution gives effect to the guidelines and to proposals in the Economic Survey. That differs from the guidelines’ illustrative chair, the Principal Scientific Adviser, and the name added “Economic”.
  • MeitY’s Digital India site records that the ministry constituted the Technology and Policy Expert Committee as a standing advisory body for that group. Tech Observer, citing the notification, reports that the MeitY Secretary chairs it and that Ravindran and Prof. Rajat Moona of IIT Gandhinagar sit on it with representatives of Nasscom, DSCI and MAIT.
  • On 10 February 2026 MeitY notified the IT Amendment Rules 2026, in force from 20 February 2026. They define “synthetically generated information”, meaning realistic audio, visual or audio-visual content made or altered by computer, and require prominent labels, according to Business Today and Khaitan & Co. This is one use of the existing IT Act route the guidelines describe.

On copyright, DPIIT’s committee issued Part I of a working paper on 8 December 2025. According to Storyboard18 and Mondaq, it proposes a hybrid model with a mandatory blanket licence for training and royalties due once an AI system is commercialised; Storyboard18 adds that MeitY supported the model and Nasscom dissented, asking for a text and data mining exception with opt-out. It is a working paper for comment, and the pages opened for this article show no final decision as of 12 November 2026.

The guidelines themselves are open about what they leave unsettled. They record that the committee “awaits the DPIIT committee’s detailed recommendations” on copyright, and they say MeitY may publish a schedule for the compliance measures in the next 9 to 12 months. For how another large jurisdiction took the opposite route, with binding risk tiers, see tuput’s article on the EU AI Act and its dates. For the related election and deepfake rules, see AI fakes in elections.

Share
Copied!

Sources & further reading

  1. India AI Governance Guidelines: Enabling Safe and Trusted AI Innovation (the full document, MeitY and IndiaAI Mission, hosted by PIB)
  2. PIB, Ministry of Electronics and IT: MeitY Unveils India AI Governance Guidelines under IndiaAI Mission (release of 5 November 2025)
  3. Reserve Bank of India: Report of the Committee to develop a Framework for Responsible and Ethical Enablement of Artificial Intelligence (FREE-AI) in the Financial Sector (press release of 13 August 2025)
  4. Digital India, MeitY: Government Constitutes Technology and Policy Expert Committee (TPEC) to Provide Expert Advisory Support to India's AI Governance Architecture
  5. The Tribune (ANI), 5 November 2025: MeitY unveils India AI governance guidelines to ensure safe, inclusive and responsible AI adoption
  6. Indian Infrastructure, 7 November 2025: MeitY unveils India AI Governance Guidelines under IndiaAI mission
  7. Construction World: Government Constitutes AI Governance and Economic Group
  8. Tech Observer, 20 April 2026: MeitY Forms Expert Panel to Guide India's AI Governance Framework
  9. Business Today, 10 February 2026: Govt tightens rules on AI-generated content, mandates labels and faster takedowns
  10. Khaitan & Co, 16 February 2026: MeitY notifies the IT Amendment Rules 2026
  11. Storyboard18: DPIIT releases working paper on copyright and AI, proposes hybrid licensing model
  12. Mondaq, 15 December 2025: Exploring the DPIIT's working paper on generative AI and copyright

Researched and written with the help of AI tools and edited for accuracy. Provided for general information and discussion only, not professional advice. See our editorial standards and disclaimer. Spotted an error? Tell us.

#india ai governance guidelines#meity#indiaai mission#ai regulation#ai safety institute#deepfakes#ai policy india

Enjoyed this? Get the next one.

One good read at a time, straight to your inbox. No spam, unsubscribe anytime.

More in AI & Technology
South Korea's AI Basic Act Took Effect on 22 January 2026 With Labelling Rules, a Safety Duty for the Largest Models and a Grace Period of at Least One Year Before Fines
What Korea's AI law asks of companies, what it offers them, which parts are paused until at least January 2027, and how India's approach compares.
The IndiaAI Mission Has Over 38,000 GPUs at Rs 65 an Hour, 12 Funded Model Builders and Open Models From Sarvam Trained on Mission Compute, With 20,000 More GPUs Announced in February 2026
Compute is priced at Rs 65 an hour, the first models are out and the February summit drew more than 100 countries. What the government's own papers say about each pillar of the mission, and the views of Carnegie India, Takshashila and an IISc-linked research head.
Sarvam's 105-Billion-Parameter Model Computes With 10.3 Billion of Them for Each Token, Using the Mixture-of-Experts Design That Runs From a 1991 Paper to Mixtral and DeepSeek
A small router sends every word to a few of 128 sub-networks, so most of the model sits idle for any one word. Here is where the design came from, how Mixtral, DeepSeek and Sarvam use it, and what it does not save.
← all articles