India Parliament
Sovereign AI ArchitectureRepublic of India

The Indian
Code.

Architecting the Global South's most robust statutory barrier between Algorithmic Innovation and Civil Liberty.

ETHICS BILL2025 Framework
MEITYAI Advisories
PbDPrivacy-by-Design
INDIA AINational Mission
Constitutional Pillars

Statutory
Domains.

India's AI strategy is partitioned into five critical domains, each governed by specific ex-ante mandates and post-market oversight bodies.

01
Regulatory Sector 01

Data Privacy & Sovereignty

Active Instruments
DPDP Act 2023 (Section 10)
IndiaAI Mission Data Stack
Sectoral Health Data Policy
Definition of Scope

Encompasses the control of personal data by the Data Principal, localized storage mandates, and the prevention of digital colonialism through strict compute borders.

Access Technical Dossier
02
Regulatory Sector 02

Algorithmic Transparency

Active Instruments
MeitY AI Advisory 2024
IT Rules 2021 (AI Amendments)
E-commerce Dark Pattern Guidelines
Definition of Scope

Mandates for the explicit disclosure of synthetic content, watermarking of deepfakes, and ensuring users are aware of neural interaction points.

Access Technical Dossier
03
Regulatory Sector 03

Liability & Accountability

Active Instruments
AI Ethics Bill 2025 (Chapter 4)
Digital India Act (Proposed)
RBI Algorithmic Credit Guidelines
Definition of Scope

Defines the legal attribution of harm generated by autonomous agents and establishes the vicarious liability of Significant Data Fiduciaries.

Access Technical Dossier
04
Regulatory Sector 04

Systemic Safety & Robustness

Active Instruments
National AI Authority Vetting Protocol
Safe AI Labs (SAIL) Mandates
Critical Infrastructure Protection (NCIIPC)
Definition of Scope

Ex-ante vetting for critical-risk neural systems, mandatory red-teaming for frontier models, and infrastructure-level security protocols.

Access Technical Dossier
05
Regulatory Sector 05

Bias & Demographic Fairness

Active Instruments
Constitutional Fairness Doctrine (Art 14)
SDF Independent Audit Mandate
NITI Aayog Responsible AI Principles
Definition of Scope

Requires the auditing of training sets to ensure linguistic and demographic representation, preventing algorithmic exclusion in public and private sectors.

Access Technical Dossier
Vertical Intelligence

Sectoral
Benchmarks.

Access jurisdictional deep-dives for industry-specific AI compliance and regulatory oversight.

Primary Statute

The 2025
Bill.

A horizontal mandate establishing the National AI Authority and strict ex-ante vetting for 'Critical-Risk' neural systems.

Section 5

Risk Classification

A 4-tier risk model adapted for Indian vernacular training sets and demographic diversity.

Section 18

National AI Authority

Dedicated statutory body for model registration, safety audits, and cross-border alignment.

Section 32

Sovereign Sandbox

Regulatory relief for domestic startups to ensure innovation is not stifled by compliance.

PbD
Privacy-by-Design Mandate

Algorithmic Sovereignty.

The DPDP Act (Section 10) mandates that AI algorithms are architected with native privacy protocols. SDFs must ensure training sets do not bleed PII.

Neural Vetting

Independent audits are required to verify that model weights do not unintentionally reconstruct sensitive personal data points.

Consent Managers

Integration with the DEPA framework ensures that every bit of data in the model's training pipeline has a revocable statutory artifact.

Regulatory Intelligence BriefREF: IND/AI/2026/001

Technical
Dossier.

This comprehensive legal and technical dossier presents an authoritative analysis of India's AI regulatory architecture. It is designed for compliance officers, general counsel, regulatory affairs teams, and policy architects navigating the intersection of artificial intelligence and Indian law.

12

Statutory Instruments

5

Regulatory Bodies

4

Risk Tiers

₹250Cr

Maximum Penalty

Dossier Contents

01Regulatory Architecture & Institutional Framework
02Risk Classification System
03DPDP Act 2023: AI Provisions
04AI Ethics Bill 2025: Core Provisions
05MeitY Advisories & Guidelines
06Sectoral Regulatory Framework
07Penalty & Enforcement Matrix
08Compliance Imperatives
09Cross Border Data Flows
10Synthetic Content & Deepfake Regulations
11IndiaAI Mission & Infrastructure
12Practitioner Guidance
01

Regulatory Architecture & Institutional Framework

India's approach to AI governance reflects a distinctive philosophy that blends sovereign control with innovation facilitation. Unlike jurisdictions that have opted for horizontal AI-specific legislation from the outset, India has constructed its regulatory architecture through a layered approach. The foundation comprises existing statutes (IT Act 2000, DPDP Act 2023), upon which sector-specific guidelines and the proposed AI Ethics Bill 2025 are progressively layered.

This architecture is not accidental. It reflects a deliberate policy choice to allow regulatory learning while maintaining the flexibility to respond to technological developments. The tripartite structure that has emerged distributes authority across policy formulation, operational oversight, and sectoral enforcement, creating a system of checks that prevents regulatory capture while ensuring domain expertise informs implementation.

M

Ministry of Electronics & IT

MeitY occupies the apex position in India's AI governance hierarchy. Its mandate encompasses policy formulation, international treaty negotiations, and oversight of the IndiaAI Mission. Critically, MeitY retains the power to issue binding advisories that create immediate compliance obligations for intermediaries and AI deployers.

Policy formulation and strategic direction
IndiaAI Mission oversight (₹10,372 Cr allocation)
International alignment and treaty compliance
Binding advisory issuance under IT Act 2000
Coordination with sectoral regulators
NAA

National AI Authority

The proposed National AI Authority under the AI Ethics Bill 2025 will serve as the operational arm of AI governance. Unlike MeitY's policy role, NAA will handle day-to-day regulatory functions including model registration, safety audits, and sandbox administration. Its composition will include technical experts alongside legal and policy professionals.

Model registration and certification
Safety audit coordination and oversight
Sovereign AI Sandbox administration
Technical standards development
Inter-agency coordination
SR

Sectoral Regulators

Domain-specific regulators retain their enforcement powers for AI deployed within their jurisdictions. This creates a matrix of overlapping authority where an AI credit scoring model, for instance, must satisfy both horizontal AI requirements and RBI-specific algorithmic lending guidelines. This layered compliance creates complexity but ensures domain expertise.

RBI: Algorithmic lending, credit scoring, fraud detection
SEBI: Algorithmic trading, robo-advisory, market surveillance
IRDAI: Automated underwriting, claims processing
TRAI: Network AI, telecom fraud detection
NMC: Healthcare AI, diagnostic systems

Institutional Hierarchy

Union Cabinet / PMO
MeitY (Policy & Strategy)
National AI Authority (Operations)
RBI
SEBI
IRDAI
TRAI
NMC
NCIIPC
02

Risk Classification System

The AI Ethics Bill 2025 introduces a four-tier risk classification framework that draws inspiration from, but meaningfully departs from, the European Union's AI Act. The Indian framework incorporates factors unique to the Indian context: linguistic diversity across 22 scheduled languages, demographic heterogeneity, infrastructural constraints in rural deployment, and the constitutional imperative of substantive equality under Article 14.

Classification is not static. The Bill empowers the National AI Authority to reclassify systems based on deployment context, scale of impact, and emerging evidence of harm. An AI system classified as "Medium Risk" in urban deployment may attract "High Risk" classification when deployed in underserved communities where algorithmic failure has disproportionate consequences.

C

Critical Risk

PROHIBITED / RESTRICTED

Systems capable of causing irreversible harm to individuals, communities, or national security. Deployment requires explicit governmental authorisation and ongoing oversight. Certain applications are prohibited outright.

Included Systems
Lethal Autonomous Weapons Systems (LAWS)
Mass surveillance and social scoring systems
Subliminal manipulation technologies
Real-time biometric identification in public spaces
Compliance Burden:

Pre-deployment governmental approval, continuous monitoring, annual recertification, designated safety officer, criminal liability for violations

H

High Risk

MANDATORY REGISTRATION

Systems that significantly impact fundamental rights, economic opportunity, or physical safety. Subject to mandatory registration, conformity assessment, and ongoing monitoring requirements.

Included Systems
Biometric identification and verification
Credit scoring and financial eligibility
Healthcare diagnostics and treatment recommendations
Educational assessment and admissions
Employment screening and HR decisions
Compliance Burden:

Registration with NAA, conformity assessment, technical documentation, bias audit, human oversight mechanisms, incident reporting

M

Medium Risk

TRANSPARENCY OBLIGATIONS

Systems that interact with users or influence decisions but do not directly impact fundamental rights. Subject to transparency and disclosure requirements.

Included Systems
Content recommendation engines
Conversational AI and chatbots
Machine translation systems
Emotion recognition in non-critical contexts
Compliance Burden:

User disclosure of AI interaction, opt-out mechanisms where feasible, content labeling, basic documentation

L

Minimal Risk

NO SPECIFIC OBLIGATIONS

Systems with negligible impact on rights or safety. No AI-specific obligations, though general consumer protection and IT Act provisions continue to apply.

Included Systems
Spam filters and email classification
Video game AI and entertainment systems
Industrial process optimisation
Inventory management systems
Compliance Burden:

No AI-specific requirements; general IT Act and consumer protection laws apply

03

DPDP Act 2023: AI Provisions

The Digital Personal Data Protection Act 2023 represents India's first comprehensive data protection legislation and creates significant obligations for AI systems that process personal data. While not AI-specific, its provisions on automated decision-making, Privacy-by-Design mandates, and enhanced obligations for Significant Data Fiduciaries directly shape the operating environment for AI in India.

Section 8(8)Right to Explanation for Automated Decisions

A Data Principal has the right to "obtain information about the logic involved in a significant decision" that is based substantially or wholly on the automated processing of their personal data. This creates a qualified right to algorithmic explanation, though the statute does not prescribe the form or depth of explanation required.

Practitioner Implications
  • "Significant decision" is undefined and will require regulatory clarification or judicial interpretation
  • The qualifier "substantially" creates ambiguity for hybrid human-AI decision processes
  • Explainability infrastructure should be built into AI systems from inception
  • Trade secret protections may limit disclosure but cannot entirely negate the right
Section 10Reasonable Security Safeguards

Data Fiduciaries must implement "reasonable security safeguards" to prevent personal data breaches. For AI systems, this extends to technical measures ensuring that training data cannot be reconstructed from model weights, that inference does not leak PII, and that adversarial attacks cannot extract training data.

Practitioner Implications
  • Differential privacy and federated learning may become de facto compliance requirements
  • Model inversion attack testing should form part of security assessments
  • Training data lineage documentation is essential for breach notification
  • "Reasonable" standard will evolve with technological capability
Section 17Significant Data Fiduciary Obligations

Significant Data Fiduciaries (SDFs) face enhanced obligations including mandatory Data Protection Impact Assessments (DPIAs), appointment of a Data Protection Officer, and periodic audits. For AI deployments, this translates to algorithmic impact assessments that evaluate bias, fairness, and discriminatory outcomes.

SDF Criteria (Likely)
  • Processing data of 1 Cr+ Data Principals
  • Processing children's personal data at scale
  • Risk of significant harm from processing
  • Government-notified entities
SDF AI Obligations
  • Algorithmic Impact Assessment
  • Bias audit and remediation
  • Third-party verification for high-risk AI
  • Annual compliance reporting
04

AI Ethics Bill 2025: Core Provisions

The AI Ethics Bill 2025 represents India's first horizontal AI-specific legislation. Currently progressing through parliamentary consultation, the Bill establishes the National AI Authority, mandates risk-based classification, and creates a sovereign sandbox regime. Importantly, it introduces criminal liability for certain categories of AI-related harm, departing from the purely civil enforcement model of the DPDP Act.

Section 5

Risk Classification

Establishes the four-tier risk classification framework. Empowers NAA to issue binding classification guidance and reclassify systems based on deployment context.

Section 12

Model Registration

Mandates registration of High-Risk and Critical-Risk AI systems with the NAA prior to deployment. Registration includes technical documentation, risk assessment, and designated responsible officer.

Section 18

National AI Authority

Establishes NAA as an autonomous body with operational independence. Defines composition (7 members including technical experts), tenure, and powers.

Section 24

Human Oversight

Requires High-Risk AI systems to maintain meaningful human oversight capability. Defines circumstances where human override must be available.

Section 28

Algorithmic Audit

Mandates annual algorithmic audits for High-Risk AI deployed by SDFs. Audits must assess bias, accuracy, and demographic fairness.

Section 32

Sovereign AI Sandbox

Creates regulatory sandbox for domestic AI startups. Provides time-limited compliance relief with enhanced monitoring.

Section 38

Synthetic Content

Requires labeling and watermarking of AI-generated content. Creates intermediary liability for platforms hosting unlabeled synthetic media.

Section 45

Criminal Liability

Introduces criminal penalties for deploying Critical-Risk AI without authorisation or causing serious harm through negligent AI deployment.

05

MeitY Advisories & Guidelines

MeitY has issued a series of advisories under Section 79 of the IT Act 2000 that create immediate compliance obligations for AI deployers. These advisories, while termed "advisory," carry binding force for intermediaries and can trigger loss of safe harbour protections for non-compliance. The March 2024 AI Advisory is particularly significant.

15 MARCH 2024MeitY AI Advisory
ACTIVE
Key Requirements
  • Permission from Government of India before deploying "under-tested" or "unreliable" AI models affecting Indian users
  • Mandatory labeling of AI-generated content including deepfakes and synthetic media
  • Explicit disclosure when users interact with AI systems
  • Prohibition on AI-generated content that violates IT Rules 2021
Compliance Implications
  • Loss of safe harbour under Section 79 for non-compliance
  • Potential criminal liability under IT Act provisions
  • Reputational risk from government enforcement actions
  • Uncertainty around "under-tested" standard interpretation
20 MARCH 2024Amended Advisory
CLARIFICATION

Following industry feedback, MeitY issued a clarification limiting the governmental permission requirement to AI platforms "that are significant in scale or potential impact." This narrows the scope but introduces ambiguity around threshold determination. The advisory now also explicitly encourages voluntary labeling commitments.

06

Sectoral Regulatory Framework

Financial Services

The RBI has issued comprehensive guidelines on algorithmic lending, credit scoring, and automated underwriting. SEBI regulates algorithmic trading and robo-advisory services. IRDAI oversees AI in insurance underwriting and claims.

RBI Guidelines

Fair lending practices, model explainability, appeal mechanisms for credit denials

SEBI Framework

Algo trading registration, risk management, audit trails, kill switches

IRDAI Circular

Non-discrimination in underwriting, transparency in premium calculation

+

Healthcare

Healthcare AI faces overlapping regulation from the National Medical Commission, CDSCO (medical devices), and the proposed Digital Health Authority under the Digital Health Mission.

NMC Guidelines

AI-assisted diagnosis must be validated by registered medical practitioners

CDSCO Rules

Software as Medical Device (SaMD) classification and approval pathways

ABDM Standards

Health data interoperability, consent framework integration

📡

Telecommunications

TRAI regulates AI deployment in network management, customer service automation, and fraud detection. The Telecommunications Act 2023 introduces new provisions relevant to AI-powered network security.

Network AI

Transparency in AI-driven network management decisions affecting service quality

Fraud Detection

AI systems for spam and fraud identification must provide appeal mechanisms

🔒

Critical Infrastructure

NCIIPC (National Critical Information Infrastructure Protection Centre) oversees AI deployment in critical infrastructure including power grids, transportation, and government systems.

NCIIPC Mandate

Security clearance for AI systems in critical infrastructure, mandatory vulnerability testing

Incident Reporting

6-hour reporting window for AI-related security incidents in CII

07

Penalty & Enforcement Matrix

Violation CategoryStatutory BasisMaximum PenaltyEnforcement Body
DPDP Act Non-Compliance (General)DPDP Act 2023, Section 33₹250 CroreData Protection Board
Data Breach (SDF)DPDP Act 2023, Section 33(b)₹200 CroreData Protection Board
Children's Data ViolationDPDP Act 2023, Section 33(c)₹200 CroreData Protection Board
Failure to Register High-Risk AIAI Ethics Bill 2025, Section 12₹50 CroreNational AI Authority
Deploying Critical-Risk AI Without AuthorisationAI Ethics Bill 2025, Section 45Criminal: 3 YearsCriminal Courts
AI-Caused Serious Harm (Negligence)AI Ethics Bill 2025, Section 45Criminal: 5 YearsCriminal Courts
Deepfake Distribution (Non-Consensual)IT Act 2000, Section 66D/66E3 Years + ₹2 LakhCyber Crime Police
Loss of Safe Harbour (Intermediary)IT Act 2000, Section 79Full LiabilityCivil/Criminal Courts
Algorithmic Lending ViolationRBI Master DirectionsLicense RevocationReserve Bank of India
Medical AI Deployment Without ApprovalDrugs & Medical Devices Rules₹1 Crore + ProsecutionCDSCO
08

Compliance Imperatives

01

Model Registration

High-risk and critical-risk AI systems must be registered with the National AI Authority prior to deployment. Registration documentation includes technical specifications, risk assessments, training data provenance, and identification of a designated responsible officer with authority to halt deployment.

02

Algorithmic Audit

Significant Data Fiduciaries deploying AI must conduct annual algorithmic audits. Audits assess bias across protected categories, accuracy metrics, and demographic fairness. Third-party verification is mandatory for public sector deployments and recommended for high-risk commercial applications.

03

Synthetic Content Labeling

All AI-generated content must bear watermarks or metadata identifiers as per MeitY Advisory (March 2024). Explicit labeling of deepfakes and synthetic media is mandatory. Platforms face intermediary liability for hosting unlabeled synthetic content under IT Rules 2021.

04

Human Oversight

High-risk AI systems must maintain meaningful human oversight capability. This includes clear escalation pathways, override mechanisms, and defined circumstances where human intervention is mandatory. Fully autonomous operation is restricted for critical decisions.

05

Data Principal Rights

DPDP Act Section 8(8) creates qualified rights to explanation for significant automated decisions. AI deployers must implement infrastructure for responding to explanation requests within statutory timelines. Documentation of decision logic is essential.

06

Privacy-by-Design

Section 10 of DPDP Act mandates that data protection be built into AI systems from inception. This includes differential privacy for training, secure inference, and technical measures preventing PII reconstruction from model weights. Annual privacy audits are recommended.

09

Cross Border Data Flows

Data Localisation Landscape

While the DPDP Act 2023 permits cross-border transfers to jurisdictions notified by the Central Government, sectoral regulations create a complex overlay. The RBI's Payment Data Localisation Directive (2018) remains fully operative, requiring payment data to be stored exclusively in India. Healthcare data under the proposed Digital Health Data Management Policy faces similar constraints.

Mandatory LocalisationPayment data, critical infrastructure, government data
Conditional TransferPersonal data to notified jurisdictions only
Permitted TransferNon-personal data, anonymised datasets

Extraterritorial Application

The AI Ethics Bill 2025 extends its reach to foreign entities in two circumstances: (a) offering AI services to Indian users, and (b) processing Indian data for AI training or inference. This creates compliance obligations for global AI providers and mirrors the extraterritorial scope of GDPR.

Practical Implications
  • • Global AI SaaS providers serving India must comply
  • • Local representative appointment may be required
  • • Jurisdictional complexity for enforcement
  • • Data processing agreements must reflect Indian law
10

Synthetic Content & Deepfake Regulations

India has moved aggressively to regulate synthetic media. The MeitY Advisory (March 2024) mandates labeling of AI-generated content, while the IT Rules 2021 (as amended) create takedown obligations for platforms hosting harmful deepfakes. Criminal liability under IT Act Sections 66D and 66E can attach for non-consensual intimate imagery and identity fraud.

Labeling Requirements

  • Visible watermarking for AI-generated images and videos
  • Metadata tagging for synthetic audio
  • Disclosure when users interact with AI chatbots
  • Platform-level content authentication systems

Criminal Liability

  • IT Act 66D: Cheating by personation using AI
  • IT Act 66E: Non-consensual intimate imagery
  • IPC 499/500: Defamation via deepfakes
  • IPC 153A: Promoting enmity via synthetic content

Platform Obligations

  • 24-hour takedown for notified deepfakes
  • Content authentication infrastructure
  • Grievance redressal for synthetic content
  • Compliance reporting to MeitY
11

IndiaAI Mission & Infrastructure

₹10,372 CrUnion Budget Allocation (2024-29)

The IndiaAI Mission represents the government's flagship initiative to position India as a global AI power. Its seven pillars encompass compute infrastructure, data platforms, application development, and crucially, a "safe and trusted AI" programme that will shape the regulatory environment.

10,000+

GPU Target

7

Mission Pillars

22

Language Models

100+

Partner Institutions

Compute Infrastructure

Public-private partnership to build 10,000+ GPU compute capacity accessible to startups and researchers

IndiaAI Data Platform

Unified datasets for AI training including anonymised government data and multilingual corpora

Application Development

Funding for AI applications in agriculture, healthcare, education, and governance

FutureSkills Prime

AI skilling initiative targeting 500,000 professionals by 2029

Safe & Trusted AI

Development of AI safety standards, testing infrastructure, and regulatory sandboxes

Startup Ecosystem

Grant funding and regulatory support for AI startups through sovereign sandbox regime

12

Practitioner Guidance

“India's AI regulatory architecture is not merely a compliance exercise. It represents a fundamental reimagining of the relationship between algorithmic power and constitutional values. Practitioners who understand this philosophy will navigate the framework more effectively than those who approach it as a checklist.”

Anandaday Misshra

Founder & Managing Partner, AMLEGALS

Immediate Action Items

1

Conduct AI inventory across all business functions to identify systems requiring registration

2

Assess DPDP Act SDF classification status and implement algorithmic audit protocols

3

Review synthetic content generation capabilities and implement labeling infrastructure

4

Establish human oversight mechanisms for high-risk AI decision systems

5

Document training data provenance and implement consent verification

6

Engage with sectoral regulators to understand domain-specific requirements

Strategic Considerations

1

Position AI governance as a board-level concern, not merely a compliance function

2

Build explainability infrastructure into AI systems from inception rather than retrofitting

3

Engage proactively with the sovereign sandbox regime for regulatory learning

4

Monitor sectoral regulatory developments across RBI, SEBI, IRDAI, and NMC

5

Develop cross-functional AI governance teams combining legal, technical, and business expertise

6

Establish relationships with the emerging AI regulatory community including NAA

AMLEGALS Technical Dossier Series

Reference: IND/AI/2026/001 | Last Updated: February 2026

Access Full White Paper Library →
MeitY Statutory Archive

Guideline Record

Full Docket →
REC_AI
Parliament of India / MeitY2025-01-15

AI Ethics & Regulation Bill, 2025 (Proposed)

The definitive horizontal statute for AI. Establishes the National AI Authority and a 4-tier risk classification system, mandating ex-ante vetting for critical-risk neural architectures.

Section 12

Mandatory Registration of Frontier Models with the National AI Authority.

Section 24

Algorithmic Impact Assessments for High-Risk Systems.

Section 45

Establishment of the AI Regulatory Sandbox for MSMEs.

REC_ME
Ministry of Electronics & IT2024-03-01

MeitY AI Advisory (Labeling & Safety)

Mandates that 'under-testing' or unreliable AI models must be explicitly labeled. Imposes strict provenance markers for synthetic content (Deepfakes) to ensure election integrity.

Para 3(b)

Consent Popup requirement for under-tested AI models.

Para 4

Metadata labeling for synthetic content.

REC_DP
MeitY / Parliament2023-08-11

DPDP Act, 2023: Algorithmic PbD

Bypasses general privacy for 'Privacy-by-Design' (PbD) in AI. Section 10 mandates Significant Data Fiduciaries (SDF) to undergo independent audits of neural training sets.

Section 8

Data Quality and Accuracy in Training Sets.

Section 10(2)

Independent Data Auditor for SDF Algorithmic Verification.