← All news and guides
AI governance

India's AI Governance Guidelines: What Companies Deploying AI Should Do Now

Are India's AI Governance Guidelines binding? A November 2026 status check, what they ask of companies deploying AI, and the controls worth building now.

Mono Team ·

A tall wooden door standing open at the end of a quiet government corridor, daylight falling across a stone floor, a wooden bench against the wall.

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.

ItemStatusSource
India AI Governance GuidelinesStatus: Policy framework. Not binding. Released 5 November 2025Source: PIB release, 5 November 2025; the Guidelines, Overview and Part 2.4
A separate AI lawStatus: 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 regulatorsSource: 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 harmsSource: 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 regulateSource: The Guidelines, Part 2.6
AI Governance and Economic Group (AIGEG) and TPECStatus: Policy bodies set up in April 2026. They coordinate and advise. They are not regulatorsSource: 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?

A long, empty corridor in an Indian government building with tall wooden doors along one wall and one door open to daylight, soft afternoon light.

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.

SutraWhat the Guidelines sayWhat it looks like in a product
Trust is the FoundationWhat the Guidelines say: Without trust, innovation and adoption will stagnateWhat it looks like in a product: Honest claims, evidence you can show a buyer, no overselling of accuracy or safety
People FirstWhat the Guidelines say: Human-centric design and human oversight; humans should have final control as far as possibleWhat 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 RestraintWhat the Guidelines say: Responsible innovation should be prioritised over cautionary restraintWhat it looks like in a product: Proportionate controls, scaled to risk, not a blanket freeze on AI use
Fairness & EquityWhat the Guidelines say: Outcomes should be fair and unbiased, with no discriminationWhat it looks like in a product: Bias tests on hiring, credit and similar uses, repeated after each model change
AccountabilityWhat the Guidelines say: Responsibility follows the function performed, the risk of harm and the due diligence doneWhat it looks like in a product: A named owner for each AI system, a contract that splits duties with vendors
Understandable by DesignWhat the Guidelines say: Explanations and disclosures users and regulators can follow, as far as technically feasibleWhat it looks like in a product: AI disclosures in the interface, plain documentation of what the system does and cannot do
Safety, Resilience & SustainabilityWhat the Guidelines say: Safeguards, anomaly detection and early warning; lightweight models encouragedWhat 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 useExisting law the Guidelines point to
Using personal data without consent to train AI modelsExisting law the Guidelines point to: Digital Personal Data Protection Act, 2023, and the IT Act, 2000
Deepfake impersonationExisting law the Guidelines point to: Bharatiya Nyaya Sanhita, 2023, and the IT Act, 2000
Misleading claims about what an AI service can doExisting law the Guidelines point to: Consumer Protection Act, 2019
Discrimination in AI-assisted hiringExisting 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 permissionExisting law the Guidelines point to: Copyright Act, 1957
Security incidents in AI systems on cloud, fintech or critical infrastructureExisting 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 recommendsWhat to build or keep
Comply with all Indian laws that apply to AI systems, and demonstrate compliance when an agency or sector regulator asksWhat 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 transparencyWhat 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 timeWhat 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 confidenceWhat 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 detectionWhat 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?

A neat stack of plain closed cardboard folders tied with cotton string on a wooden desk beside an empty wire in-tray, warm window light, shallow depth of field.

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.

HorizonWhat the Guidelines listWhy it matters to you
Short termWhat 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 chainWhy it matters to you: A risk tier and a list of applicable laws could become the common language buyers use
Medium termWhat the Guidelines list: Common standards (content authentication, data integrity, fairness, cybersecurity), an operating national incidents database, law amendments, regulatory sandboxes in high-risk domainsWhy it matters to you: Standards tend to turn into questionnaire items and contract clauses
Long termWhat the Guidelines list: Ongoing review, and new laws if risks and capabilities call for themWhy 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.

Deploying AI in India? Build the controls and evidence buyers ask for.