Policy Update
Prisha Sachdeva
Background
As AI systems become embedded in healthcare, finance, education, and governance, the risks they carry algorithmic bias, deepfakes, privacy violations, and opaque decision-making grow alongside their benefits. The Safe & Trusted AI pillar was designed as the IndiaAI Mission’s governance and risk-management layer, ensuring that the capabilities built through the other six pillars develop within an accountable, rights-respecting framework.
The IndiaAI Mission was launched in March 2024 to address existing gaps in data, research, and skills, enabling AI to contribute to India’s growth. MeitY identified seven pillars: IndiaAI Compute, Foundation Models, AIKosh, the IndiaAI Application Development Initiative, FutureSkills, Startup Financing, and Safe & Trusted AI.
The mission’s overarching goal is “AI for All,” backed by an outlay of ₹10,371 crore over five years. Of this, the Safe & Trusted AI pillar has been allocated ₹20.46 crore — the smallest allocation among the seven pillars, reflecting its research-and-governance orientation rather than infrastructure or capital deployment (Ministry of Electronics and Information Technology, Lok Sabha reply, 2026).
Functioning
The pillar promotes responsible development, deployment, and adoption of AI through indigenous governance frameworks, standards, tools, and evaluation mechanisms. It operates through three interlinked components:
Research projects: Organisations and academic institutions are selected through Expressions of Interest (EoI)to develop Responsible AI tools and frameworks. Projects address bias mitigation, machine unlearning, privacy-preserving AI, explainability, deepfake detection, and AI risk assessment.
Institutional architecture: The IndiaAI Safety Institute (AISI), announced in January 2025, advances science-based research on AI safety and governance through a hub-and-spoke model engaging academia, startups, industry, and government bodies. It also oversees implementation of the EoI-selected projects. Alongside it, the India AI Governance Guidelines propose a risk-based governance framework addressing algorithmic bias, misinformation, deepfakes, and unintended societal harm, supported by institutional mechanisms including the AI Governance and Economic Group (AIGEG) and the Technology and Policy Expert Committee (TPEC).
International engagement: India participates in global AI governance processes including the Global Partnership on Artificial Intelligence (GPAI), G20, and the United Nations, positioning Global South perspectives in international norm-setting.
Performance
| Component | Status |
| Responsible AI projects approved | 13 (across educational institutions) |
| AI Centres of Excellence (AI-CoEs) | 58 being established across States/UTs |
| India Data & AI Labs | 27 established (later reported as 31) |
| EoI rounds conducted | 2 |
| AI Safety Institute | Established January 2025 |
Source – Ministry of Electronics and Information Technology.
Selected projects include:
- Saakshya — multi-agent deepfake detection framework (IIT Jodhpur & IIT Madras)
- AI Vishleshak — audio-visual forgery detection system
- Real-Time Voice Deepfake Detection System (IIT Kharagpur)
- Bias mitigation in medical imaging — responsible AI algorithms for clinical decision-making (NIT Raipur)
- Privacy-preserving & explainable AI — federated learning models (IIT Delhi with IIIT Delhi and IIT Dharwad); security applications (DIAT)
- Machine unlearning — techniques to remove sensitive or outdated information from generative foundation models (IIT Jodhpur)
International outcomes (India AI Impact Summit 2026, New Delhi — India as Chair):
- New Delhi Frontier AI Impact Commitments — voluntary commitments by leading AI developers on safe frontier-AI development
- Guidance Note on AI Governance — practical guidance for countries designing context-appropriate governance frameworks
- Trusted AI Commons — shared repository of open resources democratising access to safe and trusted AI
Impact
The pillar’s impact is most visible at the institutional and research levels. India now has a dedicated AI Safety Institute, a published governance framework, and defined oversight bodies (AIGEG, TPEC) governance capacity that did not exist before the Mission.
At the research level, the 13 approved projects have produced named, functioning tools — particularly in deepfake detection, an area of immediate public concern in India given documented incidents involving synthetic media in political and financial contexts. The distribution of these projects across IITs, NITs, and defence research institutions also builds distributed technical capacity rather than concentrating expertise in a single body.
Internationally, India’s chairing of the AI Impact Summit 2026 and the resulting deliverables suggest the pillar has strengthened India’s credibility in global AI governance forums — an indirect but strategically significant outcome for a country positioning itself as a voice for the Global South.
However, since the pillar remains in early implementation — the AI Safety Institute was established only in January 2025, and the Governance Guidelines are more recent still — outcome-level evidence is not yet available. There is no published data on whether AI-related harms have reduced, whether industry has adopted the Governance Guidelines, or whether the developed detection tools have been deployed at scale. A definitive impact assessment would therefore be premature at this stage.
Emerging Issues
- Absence of outcome metrics: Unlike the Compute pillar (tracked GPU capacity) or FutureSkills (fellowship targets), Safe & Trusted AI has no published metrics for what it is ultimately meant to achieve — reduction in deepfake incidents, industry compliance rates, or measurable improvements in AI system fairness. Activity is documented; effectiveness is not.
- Voluntary rather than binding governance: The India AI Governance Guidelines and the New Delhi Frontier AI Commitments are framed as guidance and voluntary commitments rather than enforceable regulation. Without statutory backing or compliance mechanisms, their practical influence on industry behaviour remains untested.
- Disproportionately small allocation: At ₹20.46 crore, the pillar receives roughly 0.2% of the Mission’s total outlay — less than 1/200th of the Compute pillar’s allocation(₹4,563.36 crore). Whether this is sufficient for a mandate covering national AI safety research, governance frameworks, and international engagement is a legitimate question, particularly as AI deployment accelerates across sensitive sectors.
- Research-to-deployment gap: Tools like Saakshya and the voice deepfake detection system exist as research outputs, but there is no public information on whether they have been integrated into law-enforcement workflows, platform moderation systems, or regulatory processes where they would have practical effect.
- Timing relative to AI deployment: Governance capacity is being built while models are already being deployed — foundation models launched in February 2026, compute scaled to 38,000+ GPUs, and AIKosh crossing 13,000 datasets. If safety and governance mechanisms lag behind capability deployment, risks may materialise before mitigation frameworks are operational.
Way Forward
Clear, publicly tracked outcome indicators should be established for the pillar — for instance, the number of detection tools deployed in operational settings, industry adoption rates of the Governance Guidelines, or documented reductions in AI-related harm incidents — enabling assessment beyond activity counts.
The government should consider whether the pillar’s allocation is proportionate to its mandate, particularly as AI deployment scales across healthcare, finance, and governance. A funding review aligned with the growth of the other pillars would help ensure governance capacity does not lag behind capability.
Pathways from research to deployment should be formalised — for example, structured mechanisms for integrating tools like Saakshya into law-enforcement, electoral, or platform-moderation systems, so that research outputs translate into operational safeguards.
The transition from voluntary guidelines toward enforceable standards for high-risk AI applications (healthcare diagnostics, financial decisioning, and electoral contexts) should be considered, drawing on the risk-based approach already articulated in the Governance Guidelines.
Finally, India should sustain its international engagement momentum from the AI Impact Summit 2026, using the Trusted AI Commons and Guidance Note as ongoing contributions rather than one-time summit deliverables — ensuring the pillar’s global positioning translates into continued influence on international AI norms.
References
Ministry of Electronics and Information Technology. (2026, July 27). Safe & Trusted AI Pillar under IndiaAI Mission strengthens citizen trust through secure, fair and responsible AI solutions [Press release]. Press Information Bureau. https://www.pib.gov.in/PressReleasePage.aspx?PRID=2289946®=3&lang=1
Ministry of Electronics and Information Technology. (2026). Lok Sabha Unstarred Question No. 3246 — IndiaAI Mission budget and labs data. Government of India https://sansad.in/getFile/annex/270/AU3246_wa5YLu.pdf?source=pqars
IndiaAI. (n.d.). Safe & Trusted AI. https://indiaai.gov.in/hub/safe-trusted-ai
IndiaAI. (n.d.). India AI Governance Guidelines. https://indiaai.s3.ap-south-1.amazonaws.com/docs/guidelines-governance.pdf
IndiaAI. (n.d.). Selected projects for Responsible AI themed projects under Safe & Trusted AI pillar.https://indiaai.gov.in/news/selected-projects-for-responsible-ai-themed-projects-under-safe-trusted-ai-pillar
IndiaAI. (n.d.). Call for partnerships as part of the IndiaAI Safety Institute. https://indiaai.gov.in/article/call-for-partnerships-as-part-of-the-indiaai-safety-institute
Contributor:
Prisha Sachdeva is a Research & Editorial intern at IMPRI. She’s pursuing Bachelors in psychology (Honours) from the University of Delhi. Her interest lies in cognitive science and human behavior with a focus on evidence-based policy and behavioral research.
Acknowledgement:
The author sincerely expresses gratitude to the reviewers and the editorial team for their valuable comments, constructive suggestions, and continuous guidance throughout the preparation of this article. Their insightful feedback significantly enhanced the clarity, organisation, and analytical quality of the manuscript. The author also acknowledges the support and encouragement received during the research and writing process, which contributed to the successful completion of this work.
Reviewers:
Kaustav Majumdar and Arya Gupta
Disclaimer:
All views expressed in the article belong solely to the author and not necessarily to the organisation.
Read more at IMPRI:
The Health Assurance Agenda for Viksit Odisha
Punjab Department wise Schemes Update 2026
