Quick answer
MeitY's India AI Governance Guidelines, released in November 2025 under the IndiaAI Mission, set a principle-based approach with no standalone AI law for now. They rest on seven sutras, including trust, people first and accountability, make recommendations across six pillars, propose an AI Governance Group, a Technology & Policy Expert Committee and an AI Safety Institute, and encourage voluntary frameworks, techno-legal tools, incident reporting and graded liability.
- The guidelines are policy, not statute: they create no penalties but signal where future rules will go.
- India relies on existing laws such as the IT Act, DPDP and consumer protection, amended where gaps appear.
- Seven sutras set the principles; six pillars organise the recommendations.
- Three proposed institutions: AI Governance Group, Technology & Policy Expert Committee and AI Safety Institute.
- Voluntary frameworks, techno-legal measures, grievance mechanisms and an AI incidents database are encouraged.
- Graded liability based on function and risk means obligations should track your role in the AI value chain.
Who this is for: AI startup founders and product leaders, ML engineers and platform teams, Responsible AI and trust and safety teams, Enterprise AI governance committees, Policy and public affairs teams
What are the India AI Governance Guidelines?
The India AI Governance Guidelines are MeitY's framework for governing artificial intelligence, released in November 2025 under the IndiaAI Mission. They answer the question every builder asks first, "Is India going to pass an AI Act?", with a clear not now.
The core position
- No standalone AI law for now. The guidelines take a principle-based approach.
- Use existing laws. The IT Act, the DPDP Act, consumer protection law and sectoral rules already cover many AI harms, from data misuse to unfair practices.
- Amend where gaps exist. Targeted changes, such as the IT Rules amendments on synthetically generated information, are preferred over a horizontal statute.
Why it matters to builders
Because the guidelines are not law, they do not impose penalties. But they shape how ministries and regulators will write the rules that do: they define the vocabulary (sutras, graded liability, techno-legal measures) that future amendments and sector directions are likely to use. Designing to them now reduces rework later and gives you a credible governance story for enterprise and government customers.
For the full regulatory picture, see India AI regulation in 2026.
The seven sutras and what they mean in practice
The guidelines rest on seven sutras, or guiding principles. The left column is the guideline; the right column is one practitioner reading of how to apply it in a product team.
| Sutra | Practical translation for builders |
|---|---|
| Trust is the foundation | Treat reliability, security and honesty about capabilities as product requirements, not marketing claims |
| People first | Keep humans able to understand, contest and override decisions that affect them; design human escalation |
| Innovation over restraint | Prefer proportionate controls that let useful systems ship, scaled to risk |
| Fairness and equity | Test for performance gaps across languages, regions, gender and other groups relevant to Indian users |
| Accountability | Name owners for each AI system; keep records of design decisions, evaluations and incidents |
| Understandable by design | Disclose AI use, explain outputs at an appropriate level, and document limitations |
| Safety, resilience and sustainability | Red-team, monitor in production, plan for failure, and consider resource use |
Making sutras auditable
Principles only help if they become checks. For each sutra, define one or two artefacts a reviewer can inspect, for example an evaluation report for fairness, a model card for understandability, or an incident log for accountability. That turns a values statement into evidence.
The six pillars of recommendations
The guidelines organise their recommendations across six pillars.
- Infrastructure: access to compute, datasets and foundational capabilities that Indian AI development depends on, much of it linked to the IndiaAI Mission.
- Capacity building: skills and awareness across government, industry and the public, so that AI is used and governed competently.
- Policy and regulation: the approach of relying on existing laws and amending them where needed, rather than a new statute.
- Risk mitigation: identifying and reducing harms, including through techno-legal measures, voluntary frameworks and incident reporting.
- Accountability: clarity about who is responsible for what across the AI value chain, including graded liability based on function and risk.
- Institutions: new and existing bodies to coordinate governance, provide expertise and study safety.
Which pillars affect a typical company?
For most companies building with AI, risk mitigation and accountability translate most directly into internal practice: evaluations, incident handling, documentation and contractual allocation of responsibility. Infrastructure and capacity building matter if you participate in IndiaAI programmes or sell to government. Policy and regulation and institutions tell you where future requirements may come from.
Check indiaai.gov.in for the full text and any updates.
The proposed institutions: AIGG, TPEC and AISI
The guidelines propose three bodies to coordinate AI governance.
| Body | Broad focus (summary; read the guidelines for the precise mandate) |
|---|---|
| AI Governance Group (AIGG) | Coordination of AI governance across government |
| Technology & Policy Expert Committee (TPEC) | Expert input linking technical developments with policy |
| AI Safety Institute (AISI) | Focus on AI safety, including study and evaluation of risks |
What to watch
- Outputs from these bodies are the likeliest source of future standards, evaluation methods and voluntary frameworks in India.
- Consultations offer a route for industry to shape practical requirements; following them is cheaper than reacting to final rules.
- Interaction with existing regulators: sector regulators such as RBI and SEBI continue to act within their remits. RBI's FREE-AI framework and SEBI's AI/ML rules show sectoral governance proceeding in parallel.
Their exact composition, mandates and timelines are matters for government notification. Check IndiaAI and MeitY sources for their current status rather than relying on secondary summaries.
Voluntary frameworks, techno-legal measures and incident reporting
The guidelines favour lighter-touch, practical tools over prescriptive regulation.
Voluntary frameworks
Industry codes, self-assessments and commitments that organisations adopt before any rule requires them. For builders, adopting a recognised framework early, and documenting how you meet it, is evidence of accountability.
Techno-legal approaches
Using technology to implement legal goals by design: content provenance and labelling, consent and preference systems, privacy-enhancing technologies, and automated policy enforcement. The DPDP Rules use the same idea, describing the Data Protection Board as a digital office using techno-legal measures (Rule 20).
Grievance mechanisms
Accessible routes for people to complain about AI-driven outcomes and reach a human. This aligns with the DPDP grievance duty, where Rule 14(3) caps the published response period at 90 days.
AI incidents database
A shared record of AI failures and harms, so that lessons spread across the ecosystem. Internally, keep an AI incident log covering harmful outputs, prompt injection, data leakage, bias findings and model regressions, with root cause and remediation.
Graded liability
Responsibility should depend on function and risk. A foundation model developer, an application builder and a deployer play different roles, and obligations should match. Reflect that in contracts and documentation.
How the guidelines interact with binding law
The guidelines sit on top of laws that already bind AI systems in India.
| Binding instrument | AI relevance | Detail |
|---|---|---|
| DPDP Act and Rules | Personal data in training, prompts, outputs, logs; core rules from 13 May 2027 | DPDP Act for AI systems |
| IT Rules, 2021 (amended February 2026) | Synthetically generated information, intermediary due diligence | Deepfake and synthetic content rules |
| CERT-In Directions (28 April 2022) | 6-hour incident reporting, 180-day logs in India | CERT-In rules for AI systems |
| RBI directions | AI in regulated entities; FREE-AI recommendations once implemented | RBI FREE-AI framework |
| SEBI rules | Quarterly AI/ML reporting; regulation 16C accountability | SEBI AI/ML rules |
Example: accountability in DPDP
Rule 13(3) of the DPDP Rules requires a Significant Data Fiduciary, once notified, to observe due diligence that technical measures, including algorithmic software, are not likely to pose a risk to Data Principals' rights. That is the accountability sutra expressed as a binding duty. Expect more of this pattern as existing laws are amended.
A builder's playbook for applying the guidelines
- Inventory AI systems: model, purpose, data sources, vendors, users affected, owner.
- Classify by risk and function: are you developing a model, building an application, or deploying one? What harm could it cause? This anticipates graded liability.
- Adopt the sutras as review criteria in design reviews and launch checklists.
- Evaluate before launch: accuracy, robustness, fairness across Indian languages and groups, and safety red-teaming.
- Disclose AI use to users where they interact with it, and label synthetic outputs as the notified IT Rules require.
- Provide grievance and human escalation routes.
- Log incidents internally and be ready to contribute to a shared database if one is established.
- Map binding obligations for each system, starting with DPDP readiness for 13 May 2027.
- Track institutional outputs from AIGG, TPEC and AISI, and sector circulars.
- Review annually: re-run the inventory and classification, retire unused systems, and update documentation so it matches what is actually deployed.
None of these steps is mandated by the guidelines themselves. Together they put an organisation in a strong position when sector regulators or amended laws convert the sutras into binding requirements.
Further reading on suhasbhairav.com
- BlogAI Governance in Production: Policy, Provenance, and Safe Autonomy
- BlogPrinciples-Based Governance vs Operational Controls for Responsible AI
- BlogIncident response for AI failures in production
- AI LabBatuk AI Governance Without Friction
- AI SkillsCursor Rules Template: Incident Reporting Platform with Workflows
Before 13 May 2027
Compliance checklist
Create an AI system register
List each AI system with purpose, data, vendors, affected users and owner.
MeitY guidelines: accountability
Classify systems by function and risk
Record your role in the value chain and the potential for harm.
MeitY guidelines: graded liability
Turn sutras into review criteria
Add a check per sutra to design reviews and launch approvals.
MeitY guidelines: seven sutras
Run pre-launch evaluations
Test accuracy, robustness, fairness across Indian languages and safety.
MeitY guidelines: fairness; safety
Disclose AI interactions
Tell users when they are interacting with AI and document limitations.
MeitY guidelines: understandable by design
Offer grievance and human escalation
Provide a route to complain and reach a human for AI-driven outcomes.
DPDP Rules, Rule 14(3)
Keep an AI incident log
Record harmful outputs, leaks, prompt injection and bias findings with fixes.
MeitY guidelines: incidents database
Map binding laws per system
Check DPDP, IT Rules, CERT-In and sector rules for each AI system.
DPDP Act; IT Rules; CERT-In
Answers
Frequently asked questions
What are the India AI Governance Guidelines?
What are the India AI Governance Guidelines?
They are MeitY's principle-based framework for AI governance, released in November 2025 under the IndiaAI Mission. They recommend relying on existing laws rather than a standalone AI law for now, set out seven sutras and six pillars of recommendations, and propose new institutions including an AI Safety Institute.
Are the MeitY AI guidelines legally binding?
Are the MeitY AI guidelines legally binding?
No. They are policy guidelines and do not create penalties. Binding obligations for AI systems come from laws such as the DPDP Act and Rules, the IT Act and IT Rules, CERT-In directions and sector regulators' directions. The guidelines indicate how those instruments are likely to evolve.
What are the seven sutras of India's AI governance?
What are the seven sutras of India's AI governance?
The seven sutras are: trust is the foundation; people first; innovation over restraint; fairness and equity; accountability; understandable by design; and safety, resilience and sustainability. They are guiding principles for developing and deploying AI in India rather than detailed rules.
What is the AI Safety Institute in India?
What is the AI Safety Institute in India?
The AI Safety Institute (AISI) is one of three institutions proposed in MeitY's guidelines, alongside the AI Governance Group and the Technology & Policy Expert Committee. Its focus is AI safety. Check IndiaAI and MeitY sources for its current mandate and activities.
Will India pass an AI Act?
Will India pass an AI Act?
Not under the current approach. The November 2025 guidelines recommend against a standalone AI law for now, preferring existing laws amended where gaps appear, such as the February 2026 IT Rules amendments on synthetic content. That position could change, so follow MeitY consultations and institutional outputs.
What does graded liability mean for AI companies?
What does graded liability mean for AI companies?
Graded liability means responsibility should depend on an actor's function and the risk involved. A model developer, an application builder and a deployer play different roles in the AI value chain, so their obligations should differ. Document your role and allocate responsibilities clearly in contracts.
Legal references and sources
- India AI Governance Guidelines (MeitY, IndiaAI Mission) — November 2025 (official text)
- Digital Personal Data Protection Act, 2023 — Personal data obligations for AI (official text)
- Digital Personal Data Protection Rules, 2025 — Rule 13(3) (algorithmic due diligence for SDFs) (official text)
This guide explains the law for engineering and business teams. It is not legal advice; confirm how it applies to your organisation with qualified counsel, and check the Gazette for later notifications or amendments.
About the author

Suhas Bhairav
Systems architect and applied AI practitioner who designs production AI systems for Indian and global companies, with a focus on privacy-by-design under the DPDP Act and the EU AI Act.
Related guides
- India AI & sector rulesIndia AI Regulation in 2026: The Complete LandscapeIndia regulates AI through existing laws, not a standalone AI Act: MeitY's AI Governance Guidelines, DPDP, IT Rules on synthetic content, CERT-In directions, RBI FREE-AI and SEBI rules.
- AI & personal dataDPDP Act for AI Systems: Implementation Guide for Indian CompaniesHow the DPDP Act and Rules apply to LLM apps, RAG, agents and ML pipelines in India, with the architecture controls to build before 13 May 2027.
- India AI & sector rulesIndia Deepfake & Synthetic Content Rules (IT Rules Amendments) ExplainedHow India's IT Rules amendments on synthetically generated information work: the October 2025 draft, the February 2026 final notification, platform duties and DPDP overlap.
- India AI & sector rulesRBI FREE-AI Framework: What Banks & Fintechs Must DoWhat the RBI FREE-AI committee report asks of banks, NBFCs and fintechs, what is and is not binding yet, and how it maps onto DPDP obligations from May 2027.
- India AI & sector rulesCERT-In Rules for AI Systems: Incident Reporting & LogsHow CERT-In's April 2022 directions on 6-hour incident reporting, 180-day logs in India and clock sync apply to LLM gateways, model APIs and AI agents, and how they meet DPDP.
- DPDP Rules 2025Algorithmic Due Diligence for Significant Data FiduciariesRule 13(3) requires Significant Data Fiduciaries to verify that technical measures, including algorithmic software, are not likely to pose a risk to Data Principals' rights. Here is what that means in practice.