Short answer: no. India's AI Governance Guidelines are not a law and do not bind anyone. MeitY and the IndiaAI Mission released them on 5 November 2025. They say that "at this stage, a separate law to regulate AI is not needed" and ask companies to follow existing laws, adopt voluntary measures, publish transparency reports and run a grievance channel for AI harms. That does not make AI free of rules. The laws that already bind you, such as the DPDP Act, the IT Act and consumer law, already apply to your AI features. The Guidelines also say that some voluntary measures may later become mandatory baselines, enforced by sector regulators. So the sensible move is to build the controls now, quietly, and to keep your claims accurate.
This guide is for founders, engineering leads and compliance owners at Indian companies that build or deploy AI. It explains what the Guidelines say, what binds and what does not, and what to put in place. It is not legal advice. Take legal readings to counsel.
Are India's AI Governance Guidelines binding?
No. The Guidelines are a policy framework, not a statute or a rule. They state that the approach relies on existing laws plus voluntary measures, and the text calls voluntary measures "not legally binding". Their stated features are "optionality, flexibility, adaptability, and lack of legal enforceability or punitive action". The table separates what is soft from what is hard.
| Item | Status | Source |
|---|---|---|
| India AI Governance Guidelines | Status: Policy framework. Not binding. Released 5 November 2025 | Source: PIB release, 5 November 2025; the Guidelines, Overview and Part 2.4 |
| A separate AI law | Status: Not proposed at this stage. The Guidelines say it is not needed "given the current assessment of risks" | Source: The Guidelines, Overview, point 5 |
| Voluntary measures (codes, standards, self-certification) | Status: Optional. Some baseline measures may later become mandatory and be enforced by sector regulators | Source: The Guidelines, Overview and Part 2.4 |
| Existing laws (DPDP Act, IT Act, consumer and criminal law, copyright) | Status: Binding. The Guidelines say they already cover many AI harms | Source: The Guidelines, Part 2.3 and Annexure 4 |
| Sector regulators (RBI, SEBI, TRAI, ICMR and others) | Status: Keep their enforcement powers. Their own AI rules can bind the firms they regulate | Source: The Guidelines, Part 2.6 |
| AI Governance and Economic Group (AIGEG) and TPEC | Status: Policy bodies set up in April 2026. They coordinate and advise. They are not regulators | Source: PIB releases, 16 and 18 April 2026 |
Treat the Guidelines as a signal of where regulators and big customers are heading, not as a checklist you must pass. Do not call yourself "compliant with the India AI Governance Guidelines". There is nothing to be compliant with and nobody who certifies it.
What do the seven sutras and six pillars say?

The Guidelines are built in four parts: seven principles (sutras), recommendations across six pillars, an action plan with short, medium and long-term steps, and practical guidelines for industry and regulators. They say the sutras are "adapted from the RBI's FREE-AI Committee report" and made cross-sector. The table puts each sutra in engineering terms. The right-hand column is our reading.
| Sutra | What the Guidelines say | What it looks like in a product |
|---|---|---|
| Trust is the Foundation | What the Guidelines say: Without trust, innovation and adoption will stagnate | What it looks like in a product: Honest claims, evidence you can show a buyer, no overselling of accuracy or safety |
| People First | What the Guidelines say: Human-centric design and human oversight; humans should have final control as far as possible | What it looks like in a product: A human review or override at decisions that affect people, and a way to stop the system |
| Innovation over Restraint | What the Guidelines say: Responsible innovation should be prioritised over cautionary restraint | What it looks like in a product: Proportionate controls, scaled to risk, not a blanket freeze on AI use |
| Fairness & Equity | What the Guidelines say: Outcomes should be fair and unbiased, with no discrimination | What it looks like in a product: Bias tests on hiring, credit and similar uses, repeated after each model change |
| Accountability | What the Guidelines say: Responsibility follows the function performed, the risk of harm and the due diligence done | What it looks like in a product: A named owner for each AI system, a contract that splits duties with vendors |
| Understandable by Design | What the Guidelines say: Explanations and disclosures users and regulators can follow, as far as technically feasible | What it looks like in a product: AI disclosures in the interface, plain documentation of what the system does and cannot do |
| Safety, Resilience & Sustainability | What the Guidelines say: Safeguards, anomaly detection and early warning; lightweight models encouraged | What it looks like in a product: Monitoring, alerts, fallbacks, and a habit of using the smallest model that does the job |
The six pillars are Infrastructure, Capacity Building, Policy and Regulation, Risk Mitigation, Accountability and Institutions. Infrastructure and Capacity Building are about government programmes such as compute access and skilling. The other four shape what a deploying company will feel.
Which laws already apply to your AI features?
The Guidelines lean on existing law. Part 2.3 says many AI risks "can be addressed through existing laws", and Annexure 4 maps harms to statutes. For a company deploying AI, these are the ones that matter most.
| Harm or use | Existing law the Guidelines point to |
|---|---|
| Using personal data without consent to train AI models | Existing law the Guidelines point to: Digital Personal Data Protection Act, 2023, and the IT Act, 2000 |
| Deepfake impersonation | Existing law the Guidelines point to: Bharatiya Nyaya Sanhita, 2023, and the IT Act, 2000 |
| Misleading claims about what an AI service can do | Existing law the Guidelines point to: Consumer Protection Act, 2019 |
| Discrimination in AI-assisted hiring | Existing law the Guidelines point to: Sector and workplace laws listed in Annexure 4, including the Rights of Persons with Disabilities Act, 2016 |
| Using copyrighted work in AI content without permission | Existing law the Guidelines point to: Copyright Act, 1957 |
| Security incidents in AI systems on cloud, fintech or critical infrastructure | Existing law the Guidelines point to: CERT-In Directions, 2022: report within six hours and keep logs for 180 days |
The DPDP Act is the one most Indian AI teams will meet first. Your prompts, chat logs, embeddings and retrieved documents often hold personal data. Rule 4 and the Consent Manager registration framework have been in force since 13 November 2026. The core duties, including notice, consent, rights, security safeguards under Rule 6, breach reporting and retention, start on 13 May 2027. Our guide to DPDP-compliant AI architecture shows where it sits in an AI stack. The Guidelines themselves list open DPDP questions for AI, such as training on publicly available personal data and how purpose limitation fits generative systems, and say a detailed review is needed.
There is also the February 2026 IT Rules amendment on synthetically generated information, which adds labelling and takedown duties for AI-generated audio and visual content on intermediaries. If you generate or host such content, ask counsel how it applies to you.
What does Part 4 ask companies to do?
Part 4 is the part aimed at you. It says anyone developing or deploying AI in India "should be guided by" six points. We map them to a build and to the evidence a buyer or regulator could ask for, with the first two (comply, and demonstrate compliance) in one row.
| What Part 4 recommends | What to build or keep |
|---|---|
| Comply with all Indian laws that apply to AI systems, and demonstrate compliance when an agency or sector regulator asks | What to build or keep: A register of AI systems against the laws that touch them, plus a folder of evidence per system: data map, notices, logs, test results |
| Adopt voluntary measures on privacy and security, fairness, inclusivity, non-discrimination and transparency | What to build or keep: A short written AI policy, a risk assessment per use case and controls that match the risk. Frameworks such as ISO/IEC 42001 and the NIST AI RMF are possible anchors |
| Create a grievance redressal mechanism for reporting AI-related harms, with resolution in a reasonable time | What to build or keep: A visible contact route, a tag for AI complaints, an owner, a response target and a log of outcomes |
| Publish transparency reports on the risk of harm in the Indian context. Share sensitive material with regulators in confidence | What to build or keep: A public trust page or a short annual note, and a fuller confidential version for regulators and enterprise buyers |
| Explore techno-legal solutions: privacy-enhancing technologies, machine unlearning, algorithmic audits, automated bias detection | What to build or keep: PII redaction before model calls, deletion that reaches prompts, logs and indexes, and recurring bias or quality tests |
| Keep human oversight at critical decision points (Part 2.4, on loss of control) | What to build or keep: Human review for consequential outputs, a kill switch or fallback, audit trails |
The last row is our addition from Part 2.4, not one of the six Part 4 points. We include it because buyers ask about it. For the wider set of questions, see our security questionnaire service.
How should you handle AI incidents and complaints?

Two of the Guidelines' recommendations affect day-to-day operations: an AI incident mechanism and a grievance channel. The Guidelines borrow the OECD's idea of an AI incident, an event where the development, use or malfunction of an AI system directly or indirectly leads to a specific harm. They propose a national, federated AI incidents database, with reporting meant to be safe: it should work "without the threat of penalties", and organisations should be encouraged to report through protocols that protect confidentiality. They also say the grievance system should be separate from that database.
As of 20 November 2026, that is a plan, not a system. The action plan puts groundwork for the database in the short term and operating it in the medium term. We have not found an operational national database or an official reporting form. Separate from this, CERT-In's six-hour reporting rule for cybersecurity incidents already applies, and under the DPDP Rules a personal data breach has its own notification duties from 13 May 2027.
What to build now, whatever the national scheme turns into:
- An AI incident log. Record what happened, which system and version, who was affected, what you did, and what changed afterwards. Include near misses.
- A severity scale that tells you when an AI failure is also a security incident or a personal data breach, so the right clock starts.
- A grievance route that real users can find, with a channel for AI-specific complaints and a named owner.
- A review loop. Feed complaints and incidents into prompt, model and policy changes, as the Guidelines suggest. Our AI incident response guide covers the playbook.
Who oversees AI in India now: AIGEG, TPEC, the AI Safety Institute and sector regulators?
The Guidelines proposed an AI Governance Group, supported by a Technology and Policy Expert Committee (TPEC), with the AI Safety Institute (AISI) giving technical advice and sector regulators keeping enforcement. Two of those bodies were constituted in April 2026, one under a different name.
- AIGEG. On 16 April 2026 MeitY constituted the AI Governance and Economic Group, an inter-ministerial body chaired by the Minister of Electronics and IT. PIB says it gives formal effect to the Guidelines' institutional recommendations. It coordinates policy.
- TPEC. On 18 April 2026 MeitY constituted the Technology and Policy Expert Committee, chaired by the MeitY Secretary, to advise the AIGEG.
- AISI. The Guidelines describe the AI Safety Institute as the body for research, risk assessment, draft standards and safety testing. It supports policy. It is not an enforcement body.
- Sector regulators. RBI, SEBI, IRDAI, TRAI, ICMR and others keep enforcement powers in their domains. The Guidelines say the Finance Ministry and RBI would implement AI governance in the financial sector.
For a company, the practical point is that binding AI-specific rules are more likely to come from your sector regulator than from a new central AI law. If you sell to banks, read RBI's FREE-AI framework and the draft model risk management guidance. Both are the kind of sector material the Guidelines point to.
What is expected to come next?
The action plan lists steps over three time horizons. These are plans, and none of them is law yet. We have not seen a final text for any of the items below as of 20 November 2026.
| Horizon | What the Guidelines list | Why it matters to you |
|---|---|---|
| Short term | What the Guidelines list: India-specific AI risk classification, a regulatory gap review, voluntary frameworks, a master circular listing applicable regulations, groundwork for incidents and grievance systems, clearer liability across the AI value chain | Why it matters to you: A risk tier and a list of applicable laws could become the common language buyers use |
| Medium term | What the Guidelines list: Common standards (content authentication, data integrity, fairness, cybersecurity), an operating national incidents database, law amendments, regulatory sandboxes in high-risk domains | Why it matters to you: Standards tend to turn into questionnaire items and contract clauses |
| Long term | What the Guidelines list: Ongoing review, and new laws if risks and capabilities call for them | Why it matters to you: The no-new-law stance is for "at this stage", not forever |
Two items are worth watching. First, liability. The Guidelines support a graded liability system, where responsibility follows function, risk and due diligence, and say the IT Act should be amended to make clear how AI systems are classified, what their obligations are and how liability is imposed. Second, the text says MeitY "may publish a schedule" for compliance with these measures within 9 to 12 months. We have not found such a schedule.
How do the Guidelines compare with RBI FREE-AI and the EU AI Act?
The seven sutras come from RBI's FREE-AI report, so a team that has read one will recognise the other. The difference is reach. FREE-AI is a committee report for regulated financial entities, and RBI's model risk draft is where binding text may emerge. The India Guidelines are cross-sector and general. If you serve NBFCs or banks, sector material will bite first, as our board AI policy for NBFCs post shows.
The EU AI Act is different in kind. It is binding law with its own obligations, while India has chosen not to pass an AI law for now. An Indian SaaS company with EU customers cannot rely on India's lighter stance for those customers. See our EU AI Act guide for Indian SaaS.
What should you do now?
- List every AI system and AI feature you run or sell. Note its owner, the model and provider, the data it touches, and who it affects. You cannot govern what you have not listed. A shadow AI audit finds the tools your teams use without telling you.
- Rate each use by harm. A chatbot that answers product questions sits lower than a system that scores people or influences credit, hiring or health decisions. Write down why.
- Map personal data in prompts, logs and indexes, and plan for the DPDP start date of 13 May 2027.
- Put human oversight where it matters, with a stop switch or fallback and an audit trail.
- Set up an AI incident log and a grievance route. Name an owner and test it.
- Write a short AI policy and a transparency note. Say what you use AI for, what data you use, and who to contact. Keep it true.
- Check your contracts. Know what each AI vendor does with your data, where it processes it and what it retains, and make sure your own customer terms match.
- Watch your sector regulator for binding AI rules.
If a buyer asks about "alignment with India's AI Governance Guidelines", answer with the evidence you hold: the system register, risk ratings, oversight design, incident log, grievance route and policy. Do not say "certified" or "compliant". Mono is not a certification body and does not certify anyone. We help teams build the controls and the evidence, as part of AI governance work.
Sources: Ministry of Electronics and Information Technology, India AI Governance Guidelines, November 2025 (Overview; Parts 1 to 4; Annexures 3 to 5). PIB, "MeitY unveils India AI Governance Guidelines", 5 November 2025 (PRID 2186639). PIB, "Government constitutes AI Governance and Economic Group (AIGEG)", 16 April 2026 (PRID 2252739). PIB, "Government constitutes Technology and Policy Expert Committee (TPEC)", 18 April 2026 (PRID 2253322). Digital Personal Data Protection Act, 2023, and Rules, 2025. CERT-In Directions of 28 April 2022 as summarised in Annexure 3 of the Guidelines. Information Technology (Intermediary Guidelines and Digital Media Ethics Code) Amendment Rules, 2026, as reported by law firms. Last reviewed 20 November 2026.
FAQ
Are India's AI Governance Guidelines mandatory?
No. The India AI Governance Guidelines, released by MeitY on 5 November 2025, are a policy framework and not a law. They say a separate AI law is not needed at this stage and rely on existing laws plus voluntary measures, which they describe as not legally binding. Existing laws such as the DPDP Act and the IT Act still apply to AI systems.
Does India have an AI law?
Not as of 20 November 2026. The Guidelines conclude that a separate law to regulate AI is not needed at this stage, given the current assessment of risks. They recommend a review of existing laws for gaps and targeted amendments, for example to the IT Act on how AI systems are classified and how liability is shared. They also say new laws may follow in the long term if risks require it.
What are the seven sutras in India's AI Governance Guidelines?
They 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 they are adapted from the RBI's FREE-AI Committee report and made applicable across sectors. They guide policy, and they are not compliance requirements for companies.
What should a company deploying AI in India do about the Guidelines?
Follow the laws that already apply, then adopt the voluntary measures Part 4 suggests. In practice that means an AI system register, a risk rating per use, human oversight at consequential decisions, a grievance route for AI harms, an incident log, a short transparency note and a DPDP data map covering prompts and logs. Keep claims accurate, because there is no certificate for the Guidelines.
Who enforces the India AI Governance Guidelines?
Nobody enforces the Guidelines themselves, since they are not binding. MeitY constituted the AI Governance and Economic Group on 16 April 2026 and the Technology and Policy Expert Committee on 18 April 2026 to coordinate and advise on policy. Sector regulators such as RBI and SEBI keep their enforcement powers, and existing laws are enforced by the bodies that administer them, such as the Data Protection Board for the DPDP Act.
Can voluntary AI measures become mandatory in India?
The Guidelines say they can. They state that as the industry matures, some voluntary measures may be converted into mandatory baseline requirements enforced by sectoral regulators. No such conversion had been announced as of 20 November 2026. The likeliest route is through sector regulators, such as RBI for financial entities, rather than a single AI statute.
