Artificial Intelligence

1. Introduction: The Algorithmic Age and India's Strategic Imperative

Artificial Intelligence (AI) is no longer a speculative frontier technology; it has matured into a foundational structuring force in global economic and geopolitical competition. It is redefining how nation-states conduct warfare, manage public welfare, secure cyberspace, and shape national security. The shift from industrial to algorithmic conflict and governance necessitates a comprehensive understanding of AI's technical underpinnings, its integration into Digital Public Infrastructure (DPI), and the evolving domestic and international regulatory regimes.

For India, a nation characterized by vast socio-economic diversity, linguistic heterogeneity, and an expansive digital ecosystem, AI presents both a profound opportunity and a strategic necessity. With the launch of the National Strategy for Artificial Intelligence in 2018 under the overarching theme of #AIforAll, and the subsequent establishment of the IndiaAI Mission in 2024, India has uniquely positioned itself in the global discourse. Rather than viewing AI merely as a tool for industrial competitiveness or subscribing exclusively to existential risk narratives, India leverages AI to catalyze inclusive rural development, enhance agricultural productivity, and deliver precision healthcare, thereby linking national technological capability directly to distributive justice and welfare objectives. State intervention is designed to ensure that AI does not exacerbate existing digital divides but rather acts as a systemic equalizer.

2. Basics of Artificial Intelligence (Science & Technology)

To accurately navigate the analytical and policy aspects of the technology, one must first deconstruct its technical taxonomy. The terms Artificial Intelligence, Machine Learning (ML), Deep Learning (DL), and Generative AI (GenAI) are frequently conflated in public discourse, but they represent a precise, nested hierarchy of computational capabilities.

2.1 The Nested Hierarchy of AI

The technological evolution of AI can be conceptualized as a series of concentric circles, with each subsequent innovation driving deeper autonomy and pattern recognition.

2.2 Core Machine Learning Paradigms

Within the Machine Learning subset, algorithms are generally trained using three primary methodologies, depending on the nature of the data and the desired outcome:

• Supervised Learning: The algorithm is trained on carefully labeled datasets. The system learns the relationship between the input data and the corresponding output label, making it highly effective for classification tasks (e.g., distinguishing between a malignant and benign tumor based on historical labeled scans).
• Unsupervised Learning: The algorithm is fed vast amounts of unlabeled data and tasked with discovering hidden patterns, structures, or anomalies on its own. This is crucial for clustering consumer behavior or detecting novel cybersecurity threats.
• Reinforcement Learning: The system learns through trial and error in an interactive environment, utilizing a reward-and-punishment mechanism to optimize its decision-making. This paradigm is heavily utilized in robotics and complex game-playing AI (e.g., AlphaGo).

2.3 Generative AI Architectures and Capabilities

Generative AI relies heavily on advanced Deep Learning architectures to process and generate content. Key models include:

• Large Language Models (LLMs): These systems process human language by probabilistically predicting the next most likely word or token in a sequence, enabling highly fluent text generation and summarization.
• Generative Adversarial Networks (GANs): A dual-network architecture where a 'generator' creates synthetic content (like images) and a 'discriminator' evaluates it against real data, forcing the generator to progressively produce highly realistic outputs, often used in deepfakes.

While immensely powerful, Generative AI is prone to "hallucinations"—instances where the model confidently outputs factually incorrect or nonsensical information because it lacks a true understanding of the real world, relying solely on statistical word correlations.

2.4 Classification Based on Cognitive Capabilities (Theory of AI)

From a theoretical and evolutionary standpoint, AI is classified into four stages, highlighting the limits of current technology:
1. Reactive Machines: Systems with no memory, reacting only to current inputs (e.g., IBM's Deep Blue defeating Garry Kasparov).
2. Limited Memory AI: Systems that use recent past data to inform temporary decisions but do not retain long-term conceptual memory (e.g., self-driving cars, current LLMs within a session).
3. Theory of Mind AI: A future developmental stage where AI can understand human emotions, beliefs, and societal intentions. Currently isolated to early research phases.
4. Self-Aware AI (Artificial Super Intelligence): Hypothetical systems possessing human-like consciousness, self-awareness, and independent agency.

Currently, all deployed AI systems, including the most advanced Generative AI platforms, operate purely as Artificial Narrow Intelligence (ANI). They excel at highly specific, bounded tasks but entirely lack general reasoning or sentience.

3. The Computational Backbone: Supercomputing in India

The capabilities of Artificial Intelligence are intrinsically tied to computational power ("compute"), primarily driven by advanced Graphics Processing Units (GPUs). Control over compute infrastructure is increasingly viewed as a matter of national security and economic sovereignty, dictating the pace at which a nation can train and deploy advanced foundation models.

3.1 Historical Context and the National Supercomputing Mission (NSM)

India's supercomputing journey was catalyzed by strategic denial. In the late 1980s, after the United States refused to sell India a Cray X-MP supercomputer due to dual-use technology concerns, India established the Centre for Development of Advanced Computing (C-DAC). This led to the creation of the PARAM 8000 in 1991, India's first indigenous supercomputer.

To systematically scale these capabilities, the Government of India launched the National Supercomputing Mission (NSM) in 2015. Jointly steered by the Department of Science and Technology (DST) and the Ministry of Electronics and Information Technology (MeitY), the NSM targets the deployment of 90 petaflops of high-performance computing (HPC) capacity and complete indigenization of the supply chain by 2030.

• Infrastructure Rollout: By early 2025, the NSM had successfully deployed 34 supercomputers across the country with a combined computational capacity of 35 PetaFLOPS, directly benefiting over 10,000 researchers and 1,700 PhD scholars.
• Indigenization Milestones: The mission progressed from importing components to manufacturing indigenous HPC servers known as "Rudra" and the high-speed "Trinetra" interconnect network.

3.2 Key AI Supercomputers and Global Standing

Despite significant domestic progress, a vast capability gap remains between India's HPC infrastructure and the global frontier. The world's fastest supercomputer (Frontier, USA) operates at over 1.194 ExaFLOPS—making it approximately 90 times more powerful than AIRAWAT. India's structural vulnerability lies in its heavy dependence on foreign semiconductor manufacturing for the physical GPUs that power these systems, necessitating deep strategic investments in the domestic chip fabrication ecosystem.

4. India's Policy Architecture: National Strategy and IndiaAI Mission

India's policy approach to Artificial Intelligence avoids the binary of prioritizing unchecked commercial innovation versus heavy-handed cautionary restraint. Instead, it seeks to democratize technology as a digital public good.

4.1 The #AIforAll Strategy (2018)

In 2018, NITI Aayog released the National Strategy for Artificial Intelligence, formally establishing the #AIforAll mandate. The strategy identified five high-impact sectors for targeted AI intervention: Healthcare, Agriculture, Education, Smart Cities/Infrastructure, and Smart Mobility. The foundational premise was to position India as the "AI Garage for 40% of the World," aiming to develop highly scalable, cost-effective, and inclusive solutions that could be seamlessly replicated in other emerging economies of the Global South.

To address the severe deficit in foundational AI research, the strategy proposed a two-tiered institutional structure: Centres of Research Excellence (COREs) focusing on fundamental, theoretical AI research, and International Centers for Transformational AI (ICTAIs) dedicated to application-based research and deployment.

4.2 The IndiaAI Mission (2024-2029)

To aggressively operationalize the #AIforAll vision and transition from policy planning to systemic execution, the Union Cabinet approved the IndiaAI Mission in March 2024. Supported by a financial outlay of ₹10,371.92 crore over a five-year period, the mission is executed by MeitY and is designed to build a robust, full-stack AI ecosystem. It comprises seven core pillars:
1. IndiaAI Compute Capacity (₹4,563.36 Cr): The infrastructural bedrock of the mission. It democratizes access to processing power by onboarding over 38,000 high-end GPUs, providing them to startups, MSMEs, and researchers at a highly subsidized rate of ₹65 per hour (nearly one-third of the global average cost). The long-term target is to scale this to 100,000 GPUs.
2. IndiaAI Foundation Models (₹1,971.37 Cr): Funds the development of indigenous Large Multimodal Models (LMMs) trained on Indian datasets and languages to ensure sovereign capability and mitigate cultural bias.
3. IndiaAI Datasets Platform - AIKosh (₹199.55 Cr): Functions as a unified national data repository offering over 7,500 non-personal datasets and 273 AI models across 20 industries. By February 2026, the platform recorded nearly 70 lakh visits and over 5,000 model downloads, significantly lowering entry barriers for domestic developers.
4. IndiaAI Application Development Initiative (₹689.05 Cr): Focuses on developing AI applications for India-specific socio-economic challenges, organizing sector-specific hackathons (such as the CyberGuard AI Hackathon for cybersecurity solutions).
5. IndiaAI FutureSkills (₹882.94 Cr): Bridges the acute talent gap by funding 500 PhD fellowships, 5,000 postgraduates, and 8,000 undergraduates. It mandates the establishment of Data & AI Labs in Tier 2 and Tier 3 cities across 174 ITIs and polytechnics.
6. IndiaAI Startup Financing (₹1,942.50 Cr): Provides critical risk capital to deep-tech AI startups, facilitating not just domestic growth but global market expansion programs, such as collaborations with European innovation hubs.
7. Safe and Trusted AI (₹20.46 Cr): Focuses on developing indigenous tools for algorithmic auditing, deepfake detection, machine unlearning, and privacy-preserving architectures to ensure responsible AI deployment.

5. Sovereign AI: The Pursuit of Digital Strategic Autonomy

As AI integration deepens into national security, critical infrastructure, and public service delivery, relying entirely on foreign AI models and infrastructure poses severe strategic risks. These include data vulnerability, geopolitical throttling (denial of access to cloud resources or advanced chips), and "Western Hallucinations"—instances where foreign models generate outputs misaligned with local cultural, legal, or social contexts.

Sovereign AI refers to the capacity of a nation to develop, train, deploy, and govern AI technologies using locally controlled infrastructure, proprietary domestic datasets, and indigenous expertise. True sovereignty requires command over the entire "five-layer cake" of AI: energy/chips, compute/cloud infrastructure, foundational models, fine-tuning mechanisms, and end-user enterprise applications.

5.1 Indigenous Foundation Models and Startups

• BharatGen: Launched as part of the IndiaAI Mission, it is India's first government-funded sovereign Large Language Model ecosystem. It is designed specifically for Indian linguistic contexts, supporting 22 official languages, integrating text, speech, and document vision to enhance public service delivery without relying on Western corporate models.
• Sarvam AI: A deep-tech Indian startup that has emerged as a cornerstone of the sovereign AI push, achieving unicorn status ($1.5 billion valuation) following heavy investments from HCLTech. Selected under the IndiaAI Mission's Innovation Centre pillar, Sarvam has developed highly capable foundational models, including Sarvam-30B and Sarvam-105B (a Mixture-of-Experts architecture), trained entirely from scratch on Indian data. Their ecosystem features Sarvam for Conversations (enterprise-grade voice AI handling over 100 million interactions in 11 languages) and Sarvam Kaze (AI-powered wearable smart glasses enabling real-time visual capture and translation).
• Domain-Specific Edge Intelligence: Recognizing the limitations of cloud dependency—especially in rural areas with poor internet connectivity—Indian startups are pioneering compact, low-latency "edge AI" models. These models operate directly on local devices, supporting on-device natural language processing and translation, thereby preserving privacy and reducing infrastructure costs.

6. Sectoral Integration: Analytical Aspects and Current Affairs

The practical efficacy of India's AI strategy is best measured by its integration into existing Digital Public Infrastructure (DPI)—such as Aadhaar, UPI, and DigiLocker. This integration allows for unprecedented population-scale scaling of digital governance and service delivery.

6.1 Agriculture and Climate Resilience

Given the preponderance of fragmented landholdings and severe climate vulnerabilities, AI is fundamentally shifting Indian agriculture from input-intensive to intelligence-driven practices.

• Hyper-Local Weather Forecasting (Mission Mausam): In May 2026, the India Meteorological Department (IMD) launched advanced AI-based weather forecasting tools. The flagship system, the Bharat Forecasting System (BharatFS), operates at an unprecedented 6-km resolution (panchayat scale), utilizing AI to predict the precise block-level advance of the southwest monsoon. Furthermore, a pilot program in Uttar Pradesh utilizes advanced AI-driven downscaling techniques to provide 1-km spatial resolution rainfall forecasts up to 10 days in advance, integrating data from Automatic Rain Gauges and Doppler Radars. This represents a paradigm shift for rain-fed agricultural planning and disaster preparedness.
• Pest and Disease Management: The National Pest Surveillance System (NPSS) integrates AI with satellite imagery, meteorological data, and soil information to issue real-time preventive advisories against climate-induced pest outbreaks, safeguarding rural income security.
• Advisory & Decision Support: Kisan e-Mitra is an AI-powered, voice-enabled virtual assistant deployed by the Ministry of Agriculture. It answers farmer queries in 11 regional languages regarding critical schemes like PM-Kisan and PMFBY, processing over 8,000 queries daily.
• Yield Estimation and Insurance: The YES-TECH (Yield Estimation System based on Technology) initiative utilizes remote sensing and AI analytics for highly accurate, scientific yield estimation, drastically reducing delays in PMFBY crop insurance claim settlements. Similarly, CROPIC uses geotagged images for transparent crop damage assessment.

6.2 Rural Governance and Digital Inclusion

AI is being structurally embedded into Panchayati Raj Institutions (PRIs) to strengthen decentralized administration and reduce bureaucratic friction.

• SabhaSaar: Launched in August 2025 by the Ministry of Panchayati Raj, SabhaSaar is an AI-enabled voice-to-text summarization tool. It automatically generates structured minutes (MoM) of Gram Sabha meetings from audio and video inputs. Integrated with the BHASHINI translation engine, it supports 14 Indian languages. By February 2026, over 1,15,115 Gram Panchayats had adopted the tool, ensuring transparent, unbiased documentation and automated tracking of local resolutions.
• BhuPRAHARI: Developed in collaboration with IIT Delhi, this geospatial AI platform monitors the creation and maintenance of rural assets under MGNREGA and the Viksit Bharat-Guarantee for Rozgar and Ajeevika Mission. It utilizes high-resolution satellite imagery to assess water availability in Amrit Sarovars, replacing inefficient manual inspections with real-time tracking.
• Multilingual Governance (BHASHINI & Adi Vaani): BHASHINI is a national AI language platform integrated into over 23 government digital services, offering real-time voice and text translation across 36+ Indian languages. Surpassing one million downloads, it dismantles literacy and linguistic barriers for rural populations. Complementing this is Adi Vaani, an AI platform specifically designed to address the unique communication challenges faced by remote tribal communities.

6.3 Healthcare: Expanding Last-Mile Accessibility

The Strategy for AI in Healthcare for India (SAHI) prioritizes democratizing care and enabling universal health coverage, particularly in resource-constrained rural areas.

• Quality Benchmarking (BODH Platform): Developed by IIT Kanpur, BODH acts as a critical benchmarking ecosystem for testing AI healthcare tools for safety, bias, and accuracy prior to their clinical deployment, ensuring patient trust.
• Maternal Health (Suman Sakhi): A state-led AI WhatsApp chatbot deployed under the National Health Mission in Madhya Pradesh. It provides maternal and newborn health information directly to families and supports frontline ASHA and ANM workers, leveraging a widely used platform for last-mile outreach.
• Diagnostics and Surgical Aid: AI models are addressing the acute shortage of specialists in Tier-2 and Tier-3 cities. Tools like Scaida BrainCT process complex neuroradiology imaging to detect head trauma, while Cough Against TB analyzes sound patterns for rapid tuberculosis screening. Surgical aids like the Virtual Cardiac Twin model patient-specific heart structures to assist in complex procedures.

6.4 Space Exploration and National Defense

• Algorithmic Warfare and Defense: The character of warfare is undergoing a profound shift from industrial conflict to algorithmic operations, characterized by drone swarms, predictive targeting, and autonomous logistics. India has responded by establishing the Defence AI Council (DAIC) and the Defence AI Project Agency (DAIPA). Operational deployments include AI-based Intrusion Detection Systems (AI-IDS) under the Comprehensive Integrated Border Management System (CIBMS) for autonomous border surveillance, and the Maritime Information Management and Analysis Centre (IMAC), which integrates satellite data with AI for enhanced Indian Ocean domain awareness.
• ISRO & Space Technology: Artificial intelligence is moving from theoretical modeling to operational reality in space missions. During the historic Chandrayaan-3 mission, AI and machine learning algorithms powered the lander's sensors and cameras for autonomous hazard avoidance, facilitating the safe soft landing on the lunar south pole. Furthermore, for the upcoming Gaganyaan human spaceflight mission, ISRO developed Vyommitra, an AI-powered "half-humanoid" robot. Vyommitra is designed to simulate human physiological functions, read instrument panels, and execute microgravity experiments during the uncrewed test flights (on the HLVM3 rocket), ensuring the safety of the crew module's Environmental Control and Life Support Systems (ECLSS) prior to manned missions.

7. AI Governance, Ethics, and Global Diplomacy (Law & IR)

The rapid proliferation of AI introduces profound systemic risks: algorithmic bias, digital exclusion, privacy violations, deepfakes, and automated misinformation. India has adopted a unique, pragmatic "techno-legal" regulatory posture, differing significantly from the heavily prescriptive compliance models of the European Union and the security-first models of the United States.

7.1 India's Domestic Regulatory Ecosystem

• The Digital Personal Data Protection (DPDP) Act, 2023: While not an AI-specific law, the DPDP Act fundamentally shapes the Indian AI landscape by strictly governing how personal data is processed. It operates on a robust consent-based regime, requiring free, informed, and clear affirmative action, deliberately avoiding broad loopholes like the GDPR's "legitimate interests" clause. Under the Act, AI providers (Data Fiduciaries) bear primary accountability for algorithmic decisions involving personal data, even when processing is outsourced to third-party vendors (Data Processors). The Act places stringent prohibitions on tracking or monitoring the behavioral data of children (defined as under 18). However, critical policy gaps remain: the Act focuses exclusively on personal data, providing no regulatory oversight for non-personal data (which forms the vast bulk of AI training datasets), nor does it provide legal remedies for structural discrimination arising from algorithmic bias. Furthermore, broad state exemptions under Section 17 raise civil liberty concerns regarding unchecked state surveillance.
• The India AI Governance Guidelines (November 2025): Rather than enacting a rigid, standalone AI Act that could stifle nascent domestic innovation, MeitY issued these guidelines to implement a "light-touch," principle-based framework, adapting existing laws (like the IT Act and Copyright Act) to emerging AI realities.
  • The Seven Sutras: The framework rests on seven ethical principles—Trust, People First, Innovation over Restraint, Fairness & Equity, Accountability, Understandable by Design, and Safety, Resilience & Sustainability. The explicit prioritization of "Innovation over Restraint" highlights India's pro-growth posture.
  • Institutional Architecture: The guidelines established three permanent governance bodies:
    • AI Governance Group (AIGG): Tasked with inter-ministerial policy coordination and integrating AI with DPI.
    • Technology & Policy Expert Committee (TPEC): Providing strategic, sector-specific regulatory recommendations.
    • AI Safety Institute (AISI): Serving as India's premier technical validation body for algorithmic auditing, deepfake detection, developing national safety standards, and conducting pre-deployment safety testing of frontier models.

7.2 Global AI Governance Paradigms

The international community is currently fragmented, oscillating between stringent regulation and permissive innovation. India has strategically positioned itself as the voice of the Global South in these debates.

• The EU AI Act (August 2024): The world's first comprehensive legal AI framework, categorizing AI into four strict tiers based on risk:
  • Unacceptable Risk (Prohibited): Outright bans on social scoring by governments, subliminal manipulation, predictive policing based solely on profiling, untargeted facial scraping from the internet, and emotion recognition in workplaces.
  • High Risk: Systems used in critical infrastructure, law enforcement, employment (CV screening), or educational admissions. Imposes mandatory human oversight, extensive logging, and strict conformity assessments.
  • Limited Risk: Systems like chatbots and deepfakes. Requires strict transparency obligations (users must be explicitly informed they are interacting with AI).
  • Minimal Risk: Spam filters and video games, which remain largely unregulated.
• Bletchley Declaration (UK, Nov 2023): Signed by 28 nations, this declaration focused heavily on the existential and severe security risks posed by highly capable "frontier AI," prioritizing containment, biosecurity, and safety mechanisms.
• GPAI New Delhi Declaration (Dec 2023): Countering the UK's hyper-focus on security, this declaration—led by India—adopted a more balanced approach. While acknowledging risks, it championed AI as a catalyst for economic growth, pushed for equitable access to compute resources for developing nations, and formally introduced AI innovation in agriculture as a new thematic priority.
• AI Seoul Summit (May 2024): Led to the "Seoul Declaration" and commitments from major tech companies (OpenAI, Google) to publish safety frameworks. It formalized the creation of an international network of state-backed AI Safety Institutes to collaborate on evaluating frontier models.
• India AI Impact Summit (Feb 2026): Held in New Delhi, this summit explicitly centered the priorities of the Global South, shifting the discourse away from elite existential risks toward immediate socio-economic rights, rural livelihoods, and the deployment of inclusive Digital Public Infrastructure.

8. Systemic Challenges and the Way Forward

Despite aggressive policy momentum, India's AI trajectory faces several structural bottlenecks that must be resolved to achieve true strategic autonomy.

• The "Data Desert" Phenomenon: High-quality, digitized historical data is exceedingly scarce in rural India. If AI models are trained exclusively on urban or Western datasets, they will inevitably exhibit algorithmic bias, potentially disenfranchising rural and marginalized communities during automated welfare disbursements.
• The "Black Box" Accountability Problem: Deep learning models inherently lack explainability. If an AI system denies a citizen a welfare benefit, assigning legal accountability across the complex, multi-vendor technology stack (cloud provider, model developer, government agency) remains legally ambiguous under current frameworks.
• The Digital Divide: The widespread deployment of AI advisory tools is hindered by basic infrastructure deficits; over 25,000 Indian villages still lack robust mobile and broadband connectivity, limiting the reach of platforms like Kisan e-Mitra.
• Hardware Vulnerability: Despite the IndiaAI compute pillar, India remains overwhelmingly dependent on imported advanced semiconductors and GPUs. Until domestic fabrication capabilities mature, the ecosystem remains highly vulnerable to geopolitical supply chain shocks and export controls.

9. Memory Tips and Frameworks for Quick Retention

• Hierarchy of AI (The "Russian Doll" Model):
  • AI (The outer doll - Any smart machine, rule-based or otherwise).
  • └── ❯ ML (The inner doll - Learns from data autonomously; no explicit coding).
  • └── ❯ DL (The inner-most doll - Uses artificial neural networks; mimics the human brain).
  • └── ❯ GenAI (The magic wand - Creates novel text, images, or code).
• The 7 Pillars of IndiaAI Mission (Mnemonic: C-F-A-A-S-S-S):
  • Compute Capacity (The GPUs)
  • Foundation Models (Sovereign LLMs)
  • AIKosh (Datasets platform)
  • Application Development (Solving local problems)
  • Startup Financing (Risk capital)
  • Skills (FutureSkills)
  • Safe & Trusted AI (Deepfake detection)
• EU AI Act Risk Tiers (Mnemonic: U-H-L-M):
  • Unacceptable (Banned - e.g., Government Social Scoring, workplace emotion tracking).
  • High (Strict Rules - e.g., HR resume screening, Law enforcement biometrics).
  • Limited (Transparency needed - e.g., Chatbots, Deepfakes).
  • Minimal (No rules - e.g., Spam filters).
• Key Indian AI Platforms by Sector:
  • Language & Inclusion: BHASHINI, BharatGen, Adi Vaani.
  • Agriculture: Kisan e-Mitra, NPSS, YES-TECH, BhuPRAHARI.
  • Panchayat/Governance: SabhaSaar, eGramSwaraj.
  • Space Exploration: Vyommitra (Gaganyaan).

10. Summary

Artificial Intelligence is fundamentally rearchitecting the operational capacity of the Indian state, transitioning from theoretical research to pervasive deployment across public service delivery. Moving past the initial global hype surrounding generative algorithms, India's strategy—anchored in the #AIforAll paradigm—demonstrates a mature, pragmatic understanding of AI as a digital public good. Rather than engaging in an unwinnable capital race for pure computational supremacy against the US or China, India is carving a unique strategic niche: integrating sovereign AI models into highly robust Digital Public Infrastructure (DPI) to address population-scale socio-economic challenges.

The establishment of the IndiaAI Mission, backed by an outlay of ₹10,371.92 crore, fundamentally tackles the bottleneck of compute affordability, democratizing access to GPUs and rich datasets (AIKosh) for domestic researchers and deep-tech startups like Sarvam AI. Sectoral deployments are already yielding tangible dividends. In agriculture, hyper-local AI weather downscaling (1-km resolution) by the IMD and predictive pest surveillance are fortifying climate resilience. In rural governance, automated tools like SabhaSaar and BHASHINI are ensuring that India's profound linguistic diversity does not equate to digital exclusion, embedding transparency into the grassroots of Panchayati Raj Institutions.

However, the path forward is fraught with techno-legal complexities, including algorithmic bias stemming from "data deserts" and acute vulnerabilities in semiconductor supply chains. The tension between fostering rapid innovation and safeguarding citizen rights necessitates agile regulation. Through the DPDP Act (2023) and the MeitY AI Governance Guidelines (2025), India has opted for a principle-based, lightweight regulatory framework overseen by the newly established AI Safety Institute (AISI). By centering the priorities of the Global South—as evidenced in the GPAI New Delhi Declaration and the 2026 AI Impact Summit—India is not only building a resilient domestic sovereign AI ecosystem but is actively shaping a multipolar, inclusive global AI governance architecture.

11. Bullet Points for Prelims Easy Recall

• Generative AI vs. Machine Learning (ML): ML analyzes data to make predictions or classifications; GenAI (a subset of Deep Learning) uses neural networks to create novel, original content (text, audio, images, code).
• AIRAWAT-PSAI: India's fastest AI supercomputer (13.17 PetaFLOPS peak), installed at C-DAC Pune under the National Supercomputing Mission. Currently outpaced globally by Exascale systems.
• IndiaAI Mission (2024): Total financial outlay of ₹10,371.92 Cr. Key features include subsidizing GPU compute cost (₹65/hour) and developing sovereign foundational models.
• AIKosh: A national platform providing over 7,500 non-personal datasets and models to researchers, facilitating domestic AI development.
• BharatGen: India's indigenous, government-funded sovereign Large Multimodal Model (LLM) supporting 22 official languages.
• Sarvam AI: Deep-tech Indian startup that launched foundational models Sarvam 30B and 105B, trained specifically on Indian data, reducing reliance on Western cloud infrastructure.
• IMD AI Forecasting (Mission Mausam): Provides block-level monsoon advance forecasts (6-km resolution) and launched a pilot 1-km high-spatial-resolution rainfall forecast in Uttar Pradesh.
• SabhaSaar: AI tool launched by the Ministry of Panchayati Raj to automatically generate Voice-to-Text minutes (MoM) of Gram Sabha meetings in 14 languages.
• BhuPRAHARI: Geospatial AI tool developed with IIT Delhi for monitoring MGNREGA and Amrit Sarovar assets using high-resolution satellite imagery.
• Vyommitra: ISRO's "half-humanoid" AI robot designed to simulate human functions on uncrewed test flights (HLVM3) for the upcoming Gaganyaan mission.
• EU AI Act (2024): The world's first comprehensive AI law. Classifies AI by risk. Strictly prohibits social scoring, subliminal manipulation, and untargeted facial scraping.
• GPAI New Delhi Declaration (2023): Emphasized a balanced approach (innovation + safety) and introduced agriculture as a new AI thematic priority, contrasting the UK Bletchley Declaration’s heavy focus on existential security risks.
• MeitY AI Governance Guidelines (2025): Employs 7 "Sutras" (principles), creates the AI Safety Institute (AISI) and the AI Governance Group (AIGG). Mandates "Innovation over Restraint" and uses existing laws (like the DPDP Act) rather than drafting a new overarching AI Act.