IndiaAI Safety Institute

Several governments set up AI safety institutes from late 2023: the UK was first to announce one, in November 2023, the United States created one inside NIST, and twelve regions including the EU, Japan and Singapore signed the Seoul Declaration calling for the creation and expansion of such institutes.¹ In India, MeitY's Additional Secretary Abhishek Singh, who also oversees the IndiaAI Mission, said in October 2024 that about Rs 20 crore earmarked under the Mission's Safe and Trusted AI pillar could be used for an Institute, with more drawn from other Mission pillars.¹

The Safe and Trusted AI pillar was one of seven in the Mission approved in March 2024 and was allotted Rs 20.46 crore of the Rs 10,371.92 crore outlay, about 0.2% of the total.¹⁰

MeitY consulted stakeholders on 7 October 2024. It shared two sets of questions in advance with companies and groups including Meta, Google, Microsoft, IBM, OpenAI, NASSCOM, Access Now and several IITs, covering the Institute's focus, structure, indigenous safety tools, partnerships and role in global safety standards. MediaNama reported that the government described the Institute as a central hub on AI safety challenges that "will not act as a regulatory body, whose directions are binding". Participants urged the government to use it to identify harms and risks that could inform future regulation, asked for interoperable systems to avoid silos, suggested a risk-assessment toolkit and a voluntary compliance toolkit for industry, and debated whether MeitY should house it, preferring "reasonable independence" and suggesting a host such as an IIT.¹

Minister Ashwini Vaishnaw announced the IndiaAI Safety Institute on 30 January 2025. As summarised by Wikipedia, it would promote domestic research and development grounded in India's social, economic, cultural and linguistic diversity and Indian datasets, operate within the Safe and Trusted pillar, and work through a hub-and-spoke model of academic institutions and private partners, with UNESCO among the international collaborators.²⁶ In January 2025 Singh said the pillar would release tools that developers and deployers could use voluntarily to test for privacy protection, randomisation and algorithmic bias, planned for the IndiaAI datasets platform, and that the first step was a framework for making all AI solutions follow Indian governance requirements. His approach relies on companies, including startups, adopting such practices voluntarily.⁵

MeitY's India AI Governance Guidelines of November 2025 describe the Institute as "recently established" and recommend that it act as the main body responsible for guiding the safe and trusted development and use of AI in India. It would be involved in research, risk assessment and capacity-building, test and evaluate AI systems for risks, advise policymakers and industry, and anchor India's participation in the International Network of AI Safety Institutes. The guidelines suggest a hub-and-spoke model with a dedicated secretariat and list its functions as:⁴

• coordinating with agencies, sectoral regulators and public bodies on AI safety;
• analysing emerging risks and regulatory gaps;
• developing draft guidelines, codes, standards, evaluation metrics and testing frameworks with regulators;
• giving practical advice on voluntary risk mitigation;
• forecasting research on online safety, privacy, data governance, labour and competition;
• promoting safety tools for bias mitigation, fairness testing and explainability through platforms, APIs and open-access tools;
• public-private partnerships on tools for law enforcement and for trust and transparency;
• training programmes on AI safety;
• representing India at forums such as the Network of AI Safety Institutes; and
• supporting the AI Governance Group and the Technology and Policy Expert Committee with risk assessments, updates on industry compliance and policy recommendations.

The guidelines also say institutions such as the Institute should issue guidance notes, model codes or master circulars on how existing laws apply to AI, that it should collaborate with law enforcement and the Technology and Policy Expert Committee on how AI incident reports feed risk frameworks, and that MeitY should provide regulatory guidance in collaboration with the Institute and CERT-In. Sector regulators keep their enforcement powers. Its recommendation is that the Institute be given "adequate resources" to conduct research, develop draft standards, evaluation metrics and test methods, and provide technical guidance to regulators and industry.⁴ A January 2026 white paper from the Office of the Principal Scientific Adviser likewise lists the Institute as the body that evaluates and tests AI systems, develops safety tools, supports capacity-building and engages internationally.⁷

Projects. The Safe and Trusted AI pillar has funded eight responsible-AI projects, which MediaNama listed in July 2026:³

Leadership. On 19 May 2026 the Digital India Corporation advertised for a Director of the IndiaAI Safety Institute, with applications due on 2 June 2026. The post is on a contract or consolidated basis in New Delhi and reports to the Chief Operating Officer of the IndiaAI Mission. The Director would oversee safety research, system evaluations and tools, guidelines and standards suited to India, the hub-and-spoke engagement model, model evaluation, red-teaming and risk-assessment programmes, and integration of safety into the Mission's other pillars. Candidates need at least 20 years of post-qualification experience in AI research, safety or technology policy, including five years in senior leadership. MediaNama reported the advertisement on 9 July 2026 and the article did not report an appointment.³

Incident reporting. The guidelines recommend a national AI-incident database that would draw on CERT-In and sectoral systems, and suggest that law enforcement collaborate with the Institute and the Technology and Policy Expert Committee on how incident reports feed risk frameworks.⁴ The sources reviewed do not show the database being set up or the Institute operating it.

Funding. The pillar's Rs 20.46 crore is about 0.2% of the Mission's outlay and about five times smaller than its overheads line; MediaNama reported in April 2026 that the small allocation signals weak prioritisation of safeguards as AI deployments accelerate.¹⁰ The Institute's own budget has not been published in the sources reviewed.³

Independence and status. Participants in the October 2024 consultation wanted "reasonable independence" and suggested hosting it outside MeitY.¹ As of 2026 the Institute is a division of a MeitY-run programme rather than a statutory body, and the November 2025 guidelines themselves note that it needs resources to perform its functions.⁴

Voluntary approach. Singh's emphasis on voluntary tools has been questioned: speakers at a related MediaNama discussion argued that self-regulation has generally failed.⁵

Outputs. The sources reviewed report no published evaluations, standards, red-team results or benchmarks from the Institute, and a Director had not been reported appointed as of July 2026.³

As of 15 October 2026 the Institute is operating as part of the IndiaAI Mission, and its leadership recruitment was the latest public step in the sources reviewed. In April 2026 MeitY constituted an AI Governance and Economic Group, chaired by the Minister of Electronics and IT, with a Technology and Policy Expert Committee to advise it; the guidelines had envisaged the Institute supporting these bodies.⁸ The regulatory backdrop has also moved: in June 2026 Minister Vaishnaw said in an interview that there was a requirement for a new AI law, and on 9 July 2026 Secretary S. Krishnan said MeitY would begin stakeholder discussions on a separate law for AI regulation and then start drafting. On 8 October 2026 Vaishnaw said that the time for AI regulation had come and that MeitY would release a consultation paper within a month, with safety as its most important theme. The sources reviewed give no draft, scope or timeline for a law, and no description of the Institute's role under it.¹¹¹²⁹