On 23 October 2026, Union Electronics and IT Minister Ashwini Vaishnaw told the NDTV World Summit that the risks associated with artificial intelligence had crossed a threshold warranting legislation. He indicated that a consultation paper inviting views from citizens, industry and other stakeholders would follow in the coming weeks.That debate has moved past whether AI needs governance. What remains open is whether India can build an accountability architecture that keeps pace with its adoption. The Stanford HAI AI Index Report 2026 helps frame that question, describing a country taking up AI at remarkable speed while lacking much of the capacity to shape the systems that influence its citizens.A nation of users
Aadhaar, UPI and the wider Digital Public Infrastructure have transformed access to essential services, and AI is now being layered onto this ecosystem.The adoption indicators in the Index are striking. A University of Melbourne survey it cites finds that more than 80 per cent of Indian employees regularly use AI at work. LinkedIn data places India first in relative AI skill penetration, at three times the global average, and generative AI use among Indian university students rose from 44 per cent in 2023 to 84 per cent in 2025.The production side presents a different picture. India accounts for 0.4 per cent of granted AI patents worldwide. South Asia, with India contributing substantially, produced 21 publicly reported AI model releases by 2025, and the Index records two state backed AI supercomputing clusters in the region against 85 in China, a count that excludes planned systems and private capacity.Indian private AI investment stood at $4.09 billion in 2025, compared with $285.9 billion in the United States. Microsoft’s $17.5 billion commitment to cloud and AI infrastructure, alongside Amazon’s $35 billion investment across its Indian businesses, with AI driven digitisation among its priorities, demonstrates the scale of foreign interest.Taken together, India’s adoption is advancing faster than its ability to develop, evaluate and govern the systems on which critical services depend. A country of India’s scale cannot remain a passive recipient of technologies whose design, limitations and governing terms it has little ability to influence.The public is asking for safeguards
The Index records that India saw the sharpest rise in AI-related nervousness of any country surveyed between 2024 and 2025, increasing by 14 percentage points, while excitement rose by only two. India was also among the few countries where more respondents prioritised protective regulation over fostering innovation, by 54 per cent to 46 per cent.A separate Ipsos survey found that 65 per cent of Indians trusted their government to regulate AI responsibly. These findings indicate that citizens expect the state to ensure that AI adoption comes with safeguards.That expectation is well founded. The Index records 362 documented AI incidents in 2025, up from 233 in 2024, and finds that leading models lose accuracy sharply when a false statement is framed as the user’s own belief. In India, linguistic diversity and uneven digital literacy amplify these weaknesses, and a misleading medical recommendation or an unfair credit assessment can affect livelihoods and rights.The missing accountability layer
The India AI Governance Guidelines released by MeitY in November 2025 set out seven guiding principles across six governance pillars, and the Government constituted the AI Governance and Economic Group in April 2026.The RBI’s FREE-AI framework guides responsible adoption in finance, while the Digital Personal Data Protection framework and the IT intermediary rules address data and synthetic content. These foundations still leave one question unresolved, which is who answers for the integrity and contextual suitability of an AI generated output delivered through a platform.Consider a bank rejecting a loan application partly on the basis of an AI generated assessment. The bank may rely on a third party platform, which in turn depends on a model built by another organisation.If the assessment is discriminatory or materially wrong, responsibility fragments across the model developer, platform provider, system integrator and deploying institution, and the affected citizen is left to navigate that chain alone. India needs an identifiable, auditable and enforceable accountability function at the platform level.The case for the AI Handler
The AI Handler doctrine, which was proposed in ORF Special Report No. 322, addresses this gap. It envisages a designated accountability and assurance function within the AI platform serving Indian users. In practice, the Handler operates through technical controls, institutional responsibility and human oversight proportionate to the consequences of the application.A human officer approving every response would be impractical at scale, and a filter on what citizens may read would compromise legitimate expression, so the doctrine envisages neither. The function should have a designated and empowered human lead, supported by technical and domain specialists, with authority to initiate reviews, escalate risks and require corrective action.The Handler would validate whether systems are tested for Indian languages, social contexts and laws. It would keep records of model versions and data sources, and communicate known limitations without presenting a confidence score as proof of correctness.Safeguards would scale with potential harm, with stronger obligations in healthcare, credit, employment, education and public services, and citizens would have a clear route to complaint, correction and human review. Primary responsibility would rest with the platform for functions under its control, while developers, integrators and deploying institutions keep defined responsibilities of their own. The Handler must never become a mechanism for shifting institutional liability onto an individual employee.Agentic AI, which plans and acts on its own, makes this more urgent, since accountability must then extend to permissions, audit trails and the ability to intervene.Where Article 14 of the EU AI Act provides for human oversight of high risk systems during use, the Handler places accountability further upstream, within the platform whose provider controls much of the system’s behaviour.Sectoral regulation needs a horizontal spine
RBI, SEBI, IRDAI, TRAI and health authorities understand the risks within their sectors, and TRAI’s draft responsible AI guidelines for telecom are already open for comments. The same foundation model may nevertheless serve a bank, a hospital, a university and a government department, and its failure modes do not respect sectoral boundaries.A horizontal Handler obligation would set common minimum requirements for platform accountability, traceability, risk assessment and incident reporting, on which sector regulators could build domain specific safeguards.Industry would gain clarity, since organisations surveyed for the Index cite regulatory uncertainty as a significant obstacle. India should build on standards such as ISO/IEC 42001 without treating certification as a substitute for independent evaluation or legal accountability, and the IndiaAI Safety Institute can develop testing methods suited to Indian requirements.Transparency should be defined carefully. Regulators need protected access to audit evidence and citizens need explanations of decisions affecting them, while proprietary source code and model weights can remain confidential.An Indian AI vision towards 2047
By 2030, India should have an enforceable, risk based AI framework in force, with platform level accountability, the Handler function, incident reporting, independent evaluation and citizen redress, coordinated with sector regulators. By 2035, it should have built sovereign compute, Indic foundation models and quality datasets, with AI integrated responsibly into Digital Public Infrastructure.By 2047, India should be shaping international AI governance standards for developing economies, building on the New Delhi Declaration adopted at the AI Impact Summit in February 2026.The forthcoming consultation is the opportunity to move beyond broad principles and settle who answers when a machine influences a consequential decision. Responsibility should rest with those who develop, provide and deploy the technology, protecting the citizen who trusted the system and the employee who had no say in its design. India built its digital infrastructure at population scale, and the intelligence now being layered onto it must be governed with the same resolve.(The author is a former National Cyber Security Coordinator, Government of India, and former Signal Officer-in-Chief of the Indian Army. Views expressed are personal)
