India Launches DEPA Framework to Enable Secure AI Training While Protecting Data Sovereignty

Bengaluru, October 31, 2026: India has introduced a new framework designed to enable artificial intelligence systems to learn from large and diverse datasets while protecting privacy, intellectual property and institutional control over sensitive information. The Ministry of Electronics and Information Technology (MeitY) launched the Handbook on Data Empowerment and Protection Architecture (DEPA) for AI Training at a high-level roundtable in Bengaluru, marking an important development in the country’s efforts to establish a secure and responsible data ecosystem for artificial intelligence.

The handbook was unveiled by MeitY Secretary S. Krishnan with support from the iSPIRT Foundation. It outlines a technology and legal framework intended to facilitate collaboration among organisations holding valuable datasets while ensuring that participating institutions retain control over their information.

The initiative addresses a fundamental challenge facing artificial intelligence development. Training advanced AI models requires access to large quantities of relevant, high-quality data, but much of this information is held by separate institutions that cannot freely share it because of privacy obligations, commercial confidentiality, intellectual property restrictions and regulatory requirements.

The DEPA for AI Training framework seeks to overcome these barriers by creating mechanisms through which organisations can collaborate on AI development without requiring unrestricted transfer of sensitive datasets.

Addressing India’s AI Training Data Challenge

Artificial intelligence systems depend on high-quality training data to identify patterns, develop analytical capabilities and generate useful predictions. The effectiveness of these systems is influenced not only by computing power and model architecture but also by the quality, diversity and relevance of the information used during training.

India possesses substantial datasets across sectors such as healthcare, banking, education, agriculture, scientific research and public administration. However, these datasets are often distributed across institutions operating under different legal, technical and administrative requirements.

A hospital may possess valuable medical imaging records, while a university may hold specialised research data and a financial institution may maintain extensive information about financial transactions. Although such information could contribute to the development of specialised AI systems, unrestricted sharing may expose individuals and organisations to privacy, security and commercial risks.

The Ministry of Electronics and Information Technology has identified these concerns as significant obstacles to wider data collaboration.

The DEPA Training framework aims to provide a structured approach through which institutions can make data available for authorised AI training activities without relinquishing control over the underlying information.

This approach could help researchers and technology developers access more representative datasets while allowing participating institutions to maintain appropriate safeguards.

A Technology and Legal Framework for Trusted Data Collaboration

The DEPA for AI Training Handbook establishes a framework combining technical systems with contractual permissions and institutional safeguards.

Rather than relying exclusively on conventional arrangements in which datasets are transferred from one organisation to another, the framework supports controlled access to data for clearly defined purposes.

Participating institutions can establish the conditions under which their information may be used, including the authorised purpose of a training activity and the permissions governing access.

Controlled computing environments are intended to help ensure that AI training operations take place within defined technical boundaries.

Privacy-enhancing safeguards are incorporated to reduce the risk of unauthorised access, disclosure or misuse of sensitive information.

The framework also introduces provisions for data provenance, enabling the origin of information, permissions associated with its use and its role in the training process to be recorded and traced.

Such mechanisms are intended to strengthen accountability and transparency when several organisations contribute information to a common AI development initiative.

The approach recognises that responsible artificial intelligence requires more than access to computing infrastructure. It also requires reliable arrangements for data governance, institutional trust and the protection of individual and organisational rights.

AI Models Can Learn Without Moving Sensitive Data

One of the central ideas behind the DEPA framework is that artificial intelligence models can be brought to the location where data is held, rather than requiring the underlying datasets to be transferred to a separate central repository.

This approach was highlighted by Professor Ajay Kumar Sood, Principal Scientific Adviser to the Government of India, during the Bengaluru roundtable.

Under such an arrangement, authorised computing processes can be carried out within controlled environments associated with the organisations holding the relevant data.

For example, multiple research institutions could potentially collaborate on the development of an AI system while maintaining their own datasets under separate institutional control.

Instead of transferring complete collections of sensitive records to another organisation, participating institutions could permit authorised training operations under agreed conditions.

This approach may reduce some of the privacy and confidentiality risks associated with conventional data-sharing arrangements.

However, secure training environments still require careful technical implementation, access controls and safeguards against the unintended disclosure of information through model outputs or other channels.

The DEPA framework is intended to establish a foundation for such collaboration rather than eliminating every risk associated with the use of sensitive data in artificial intelligence.

Data Provenance to Strengthen Transparency and Accountability

An important feature of the framework is its emphasis on data provenance.

Data provenance refers to the ability to identify where information originated, how it was obtained, what permissions apply to its use and how it has been processed.

In artificial intelligence development, these questions are particularly important because training datasets may originate from numerous institutions and contain information subject to different legal or contractual restrictions.

When data from multiple sources contributes to the development of an AI model, maintaining reliable records of its origin and authorised use can become a complex responsibility.

The DEPA framework seeks to address this challenge by incorporating mechanisms that allow relevant information about data use to be traced throughout the training process.

Such records can help participating organisations understand whether information has been used in accordance with agreed permissions.

They may also support auditing, accountability and the investigation of potential compliance concerns.

For research institutions, data provenance is relevant to scientific reproducibility and the ability to document the sources and conditions associated with research findings.

For commercial organisations, it can contribute to stronger management of intellectual property, confidential information and data-related contractual obligations.

The inclusion of provenance mechanisms reflects the importance of establishing trust among organisations participating in collaborative AI development.

Strengthening India’s Artificial Intelligence Research Ecosystem

The framework has potential implications for India’s universities, scientific laboratories, research institutions and technology companies.

Artificial intelligence research frequently requires access to specialised datasets that may not be available through conventional public information sources.

Researchers developing systems for medical diagnosis, agricultural analysis, language processing or industrial applications may benefit from information held across several institutions.

However, fragmented access to such information can limit the ability to develop robust models.

The DEPA framework offers a proposed approach through which institutions could collaborate while maintaining control over their respective datasets.

Professor Ajay Kumar Sood highlighted the potential of the architecture to support multi-institutional scientific research and accelerate discoveries that require access to information distributed across different organisations.

Such collaboration could be particularly relevant to research problems involving complex patterns that cannot be adequately studied using data from a single institution.

The framework may also create opportunities for smaller research organisations and startups to participate in collaborative AI development without independently acquiring extensive proprietary datasets.

Nevertheless, the practical benefits will depend on how institutions adopt the framework, establish governance arrangements and make authorised data access available.

Applications in Healthcare and Medical Research

Healthcare represents a potential area of application for trusted AI training because medical institutions possess extensive datasets that may be valuable for research.

These datasets can include diagnostic images, laboratory results, clinical observations and other information generated during medical treatment.

Artificial intelligence systems trained on appropriately governed medical data may assist researchers in identifying patterns associated with diseases, evaluating medical images and developing analytical tools.

However, healthcare information is particularly sensitive, and its use requires safeguards addressing patient privacy, confidentiality and applicable legal requirements.

A framework that supports controlled computing and purpose-specific permissions could enable medical institutions to collaborate on research without routinely transferring complete patient datasets to external developers.

For example, hospitals participating in a research programme could potentially permit authorised AI training activities within controlled environments while retaining responsibility for the information they hold.

Such arrangements would still require appropriate institutional approvals, security controls and compliance with applicable regulations.

The DEPA framework offers a possible foundation for these activities, although its launch should not be interpreted as confirmation that unrestricted access to medical records has been authorised.

Opportunities for Banking and Financial Technology

Financial services represent another sector where trusted data collaboration may support the development of specialised artificial intelligence applications.

Banks and other financial institutions maintain extensive information about transactions, financial products and customer relationships.

These datasets can be valuable for developing AI systems intended to support fraud detection, risk analysis and improvements in financial services.

However, financial institutions operate under strict confidentiality and regulatory requirements that limit how customer information may be shared.

During the Bengaluru roundtable, Anjani Rathor, Group Head of Digital Banking and Customer Experience at HDFC Bank, highlighted the potential of trusted data collaboration for artificial intelligence in banking.

He referred to HDFC Bank’s experience with DEPA Inference and the opportunity to build upon that experience through DEPA Training.

The distinction between inference and training is important. AI inference involves using an existing model to process information or generate an output, while AI training involves developing or improving a model using relevant data.

The new handbook focuses on establishing trusted arrangements for the latter activity.

If implemented effectively, such arrangements could help financial institutions and technology companies collaborate on AI development while preserving the contractual and regulatory protections applicable to financial information.

Supporting Indian AI Startups and Indigenous Innovation

The initiative is relevant to India’s growing community of artificial intelligence startups, particularly companies developing specialised applications for domestic industries.

Startups often face difficulties obtaining sufficiently large and representative datasets because valuable information is controlled by established institutions.

Acquiring data through commercial agreements can also involve substantial costs, while some information cannot be transferred freely because of legal or confidentiality restrictions.

A structured framework for trusted data collaboration could provide an alternative pathway for participating in AI research and development.

Under appropriate agreements, startups may be able to collaborate with institutions holding relevant datasets without requiring unrestricted possession of the underlying information.

This approach could support the development of specialised AI models for sectors such as healthcare, finance, agriculture, education and industrial operations.

It may also encourage partnerships between technology companies, universities and established businesses.

However, the handbook does not automatically grant startups access to institutional datasets. Participation will depend on the permissions, contractual arrangements and technical safeguards established by the organisations involved.

The framework therefore represents an enabling approach rather than a guarantee of universal data availability.

Building on India’s Digital Public Infrastructure Experience

The DEPA initiative reflects India’s broader experience in developing digital public infrastructure and technology frameworks intended to support secure information exchange.

India has developed several large-scale digital systems that provide common infrastructure upon which public institutions and private enterprises can build services.

The DEPA approach extends this emphasis on structured, permission-based collaboration into the area of artificial intelligence training.

During the launch, Sharad Sharma, Co-founder of the iSPIRT Foundation, highlighted the significance of the framework for India’s evolving data economy.

He connected the initiative to the country’s earlier digital infrastructure developments and emphasised the opportunities created by trusted data systems.

The underlying principle is that technological infrastructure becomes more valuable when different institutions can participate through interoperable arrangements and clearly defined responsibilities.

In the context of AI training, this means creating the conditions for authorised data collaboration without making unrestricted data transfers a prerequisite for innovation.

The success of such an approach will depend on technical interoperability, institutional participation and effective implementation.

Data Sovereignty and Institutional Control

An important objective of the DEPA framework is the preservation of institutional control over data.

Data sovereignty concerns the legal and institutional authority governing how information is stored, accessed, processed and used.

For organisations handling sensitive or commercially valuable information, maintaining control over data can be essential to protecting privacy, intellectual property and contractual obligations.

The DEPA framework seeks to enable collaboration while preserving this control through defined permissions and technical safeguards.

The approach may be relevant to collaborations involving institutions located in different jurisdictions.

Such arrangements can create opportunities for scientific research but also raise questions concerning legal compliance, data protection and the responsibilities of participating organisations.

Professor Ajay Kumar Sood highlighted the framework’s potential contribution to international scientific cooperation, including collaborations involving multiple institutions and countries.

He described the approach as an opportunity for India to contribute to emerging international discussions on responsible artificial intelligence and trusted data access.

However, any cross-border application would remain subject to applicable national laws, regulatory requirements and contractual conditions.

The handbook does not independently authorise transfers or uses of information that would otherwise be restricted under existing law.

Industry and Research Leaders Participate in Bengaluru Roundtable

The launch brought together representatives from government, technology companies, industry, academic institutions and research organisations.

Alongside MeitY Secretary S. Krishnan and Principal Scientific Adviser Professor Ajay Kumar Sood, the discussions included senior representatives from the iSPIRT Foundation and the private technology sector.

Kris Gopalakrishnan, Co-founder of Infosys and Chairman of Axilor Ventures, highlighted the opportunities for India to advance responsible artificial intelligence through trusted access to research data.

He emphasised that collaboration involving diverse datasets could help researchers develop more robust AI models while protecting the rights of participating institutions and data contributors.

The roundtable also featured a presentation on the DEPA Training Framework by Hari Subramanian.

Other participants included Manav Garg, Executive Chairman of Emergent AI, Ashwani Sharma of Sarvam and Adith Charlie, Managing Editor of YourStory.

Their participation reflected the broad relevance of data collaboration to India’s technology industry, research community and emerging artificial intelligence ecosystem.

The discussions examined the potential of trusted data arrangements to support AI innovation and applications serving wider public interests.

Implementation Will Determine the Framework’s Impact

Although the launch represents an important development in India’s approach to AI data governance, the handbook establishes a framework rather than demonstrating that a fully operational national data-sharing system has already been deployed.

Its practical impact will depend on the willingness of institutions to participate, the availability of suitable computing infrastructure and the establishment of reliable contractual and governance mechanisms.

Participating organisations will need to determine how access permissions are granted, how sensitive information is protected and how compliance can be demonstrated.

Technical safeguards will also need to address risks associated with unauthorised access, model security and the unintended disclosure of information.

The development of interoperable systems and common operational standards may help reduce the complexity of collaboration among institutions.

MeitY has expressed its commitment to taking the framework forward and supporting its application within India’s AI ecosystem.

The initiative therefore provides a foundation for further work involving government agencies, research institutions and industry participants.

Its success will ultimately depend on whether the proposed safeguards can be translated into reliable, accessible and commercially practical arrangements.

Towards a Trusted Indian Artificial Intelligence Ecosystem

The launch of the DEPA for AI Training Handbook marks an important step in India’s efforts to address the challenges of data access, privacy and institutional trust in artificial intelligence development.

By combining contractual permissions, controlled computing environments, privacy-enhancing safeguards and data provenance, the framework seeks to establish conditions under which organisations can collaborate without surrendering control over sensitive datasets.

The initiative could be relevant to research institutions, universities, hospitals, financial organisations, startups and businesses developing specialised AI applications.

It also highlights the growing recognition that access to high-quality data is as important to AI development as computing infrastructure and advanced algorithms.

For India, the ability to support responsible collaboration across institutions could strengthen domestic research capabilities and create opportunities for developing AI systems suited to the country’s languages, industries and public-service requirements.

However, the handbook’s release represents the introduction of an enabling framework, not the completion of a nationwide operational system or a replacement for existing data protection obligations.

As institutions examine its proposed arrangements, the DEPA framework may contribute to a more structured approach to data collaboration in which innovation, privacy, accountability and institutional control are addressed together.

The development reflects India’s continuing effort to build an artificial intelligence ecosystem that combines technological advancement with responsible data governance and greater domestic control over critical digital resources.

---

References

Ministry of Electronics and Information Technology, Press Information Bureau, Government of India, 31 October 2026. MeitY Launches DEPA for AI Training Handbook to Advance Trusted Data Collaboration for AI. The official announcement confirms the launch of the handbook at a high-level roundtable in Bengaluru, its development with support from the iSPIRT Foundation, the principal technical and legal safeguards, and statements from government, research and industry representatives.

Source: https://www.pib.gov.in/PressReleasePage.aspx?PRID=2330756

---