India's national security architecture is undergoing a critical transformation, necessitating a deeper understanding of "Atmanirbharta" beyond mere technology acquisition to encompass doctrinal reform, particularly as artificial intelligence increasingly influences decision-making. The article argues that true self-reliance in national security requires control over decision support systems, addressing issues of authority distribution, governance, and explainability, as dependency on foreign AI models trained on external data creates significant risks of misaligned outputs and loss of judgment. India is pursuing a five-layer Sovereign AI framework—covering applications, models, compute, infrastructure, and energy—with a focus on democratizing compute and decentralizing innovation through affordable access, leveraging its vast citizen-generated datasets for training indigenous models. However, challenges remain in ensuring central governance for decentralized capabilities, implementing legally mandated explainability and data privacy audit regimes, and achieving compute sovereignty through supply chain diversification, emphasizing that the ultimate goal is the sovereign, indigenous planning and execution of national security missions.

India's national security architecture is undergoing a critical transformation, necessitating a deeper understanding of "Atmanirbharta" beyond mere technology acquisition to encompass doctrinal reform, particularly as artificial intelligence increasingly influences decision-making. The article argues that true self-reliance in national security requires control over decision support systems, addressing issues of authority distribution, governance, and explainability, as dependency on foreign AI models trained on external data creates significant risks of misaligned outputs and loss of judgment. India is pursuing a five-layer Sovereign AI framework—covering applications, models, compute, infrastructure, and energy—with a focus on democratizing compute and decentralizing innovation through affordable access, leveraging its vast citizen-generated datasets for training indigenous models. However, challenges remain in ensuring central governance for decentralized capabilities, implementing legally mandated explainability and data privacy audit regimes, and achieving compute sovereignty through supply chain diversification, emphasizing that the ultimate goal is the sovereign, indigenous planning and execution of national security missions.

India's national security architecture is undergoing a critical transformation, necessitating a deeper understanding of "Atmanirbharta" beyond mere technology acquisition to encompass doctrinal reform, particularly as artificial intelligence increasingly influences decision-making. The article argues that true self-reliance in national security requires control over decision support systems, addressing issues of authority distribution, governance, and explainability, as dependency on foreign AI models trained on external data creates significant risks of misaligned outputs and loss of judgment. India is pursuing a five-layer Sovereign AI framework—covering applications, models, compute, infrastructure, and energy—with a focus on democratizing compute and decentralizing innovation through affordable access, leveraging its vast citizen-generated datasets for training indigenous models. However, challenges remain in ensuring central governance for decentralized capabilities, implementing legally mandated explainability and data privacy audit regimes, and achieving compute sovereignty through supply chain diversification, emphasizing that the ultimate goal is the sovereign, indigenous planning and execution of national security missions.

India’s national security architecture is transitioning through a decisive juncture. As artificial intelligence begins to dominate decision support systems, which were once solely within the realm of human judgment, the idea of Atmanirbharta demands a more rigorous understanding. Self-reliance, especially within the national security architecture, cannot remain a mere technology acquisition strategy. The time is ripe when we must evolve it into a doctrinal reform to address who and what controls state decision-making systems, how authority is distributed within them, what governance safeguards govern their use, and where the sources of origins and the point of impact of such decisions are, in a clearly explainable manner.

The general articulation of Atmanirbharta has mostly remained rooted in economic resilience and industrial capacity. While important, this framing often makes little of what is at stake when it comes to national security. Dependence here goes much beyond the hardware of national defence; it is as critically inconvenient in the decision support systems that power the targeting and firing frameworks and guidelines that add the lethality which we need as a nation when it comes to national security. When core facets like intelligence, policing, and justice-delivery are facilitated by systems that the India does not fully own or at the very least, control, then underlying sovereignty becomes provisional.

Fundamental risk of dependency

When it comes to our national security in the AI era, the challenge is not access to technology, but loss of judgment. AI models trained on external datasets, designed for foreign operational contexts, and hosted on foreign infrastructure embed assumptions that may be misaligned with India’s legal, cultural, and threat environments. The resultant output is often contextually irrelevant, and the decision tree is also opaque and beyond the scrutiny of the Indian agencies.

The risk for the nation is further accentuated in today’s geopolitical environment, where dependence is often a leverage. Reliance on foreign hyperscalers may lead to a critical chokepoint that cannot be mitigated through contractual assurances alone. National security cannot be outsourced to systems whose continuity, reliability and auditability lie beyond our sovereign control. Yet, at times, there is a degree of normalisation of this dependence by prioritising rapid adoption.

Achieving sovereignty in AI-driven national security landscape

To achieve sovereignty, different nations and blocs have followed different pathways. For instance, the European Union is putting a regulatory regime in place through its AI Act, which was passed in 2024, and key provisions of it will come into force this August. This is a rights-based regulation focusing on safety and explainability. The US, on the other hand, is building and collaborating with hyperscalers and exporting these models globally. Predictably, the Chinese are focusing on state control.

For India, the larger pathway for this Atmanirbharta has been laid out in the five-layer framework defined by the government when it comes to AI, sovereignty in applications, models, compute, infrastructure and energy. In other words, India defines Sovereign AI as a full-stack national capability across applications, models, compute, infrastructure, and energy, enabling the country to build and run AI entirely on its own terms. This has been well articulated in the recently concluded India AI Impact Summit that India hosted. India’s most impactful decision has been its policy on democratising compute and decentralising innovation. Enabling advanced computing accessible beyond metropolitan and corporate enclaves is a structurally sound reform that aligns with India’s scale and diversity. Making it affordable at ₹65 per hour is the true genius.

From a national security perspective, this could mean operational and targeting support for a Corp Commander responding to enemy aggression, analytical augmentation capabilities for and analyst in IB who is trying to make sense of an intercept in foreign language, assistance in following the due process of law to a freshly minted constable posted in the hinterlands with limited access to legal knowhow, or truly democratizing the justice delivery process by keeping citizen rights at the forefront with solutions like Nyaya Setu.

The other key strength that we possess is the elephantine database that our citizens produce, and which is key to training, contextualising, and deploying our own models. Looking through a national security lens, this could mean that erstwhile siloed data is brought together in a conjoint data lake. For instance, the world’s largest justice system ERP is India’s CCTNS-ICJS platform. The richness of this humongous dataset can be appreciated when we realise that an entry in this system acts as a single source of truth for the entire criminal justice system of the nation over the last two decades. For instance, data entered in registering a case in a village in Mizoram becomes an enabler for policymaking when it is aggregated and analysed nationally at the Union home ministry. When we add to that the NATGrid data or the enormous public-safety net of CCTVs deployed through India’s Smart City project, or the petabytes of social media chatter we produce daily, this could form a data lake which can be leveraged to generate substantial predictive capabilities.

Challenges ahead

Decentralised capability without central governance only defuses risk. While privacy preservation governance and legal safeguards are being put in place with laws like DPDP Act or the revision of IT Rules, given the speed of development in these technologies, there needs to be legally mandated explainability embedded into applications and models to root out biases quickly. As the country comes together to respond to this national mission, it is equally important that we work towards achieving a certain degree of sovereignty in computing through indigenisation and diversification of supply chains. Unless we put these guardrails in place, the stride of innovation could potentially outpace the administrative controls in place.

Additionally, while in theory, our sovereign datasets present unprecedented analytical potential, these data could potentially usher in dramatic improvements in predictive policing, threat assessment, and institutional learning.

In practice, however, they are limited by institutional silos and governance opacity. Data integration has progressed faster than clarity on purpose limitation, access controls, and administrative oversight. Without enforceable governance structures, data consolidation could potentially become a tool that would fall short of delivering strategic effectiveness.
The reform challenge, therefore, is less of a technical issue than an institutional and administrative one. India needs legally mandated data privacy audit regimes that demonstrate at a predefined periodicity as to who is accessing what, for what purpose, and under whose supervision. Thus, reform must move towards mandatory explainability for all state-deployed AI systems, especially those influencing national security decision-making.

Compute sovereignty remains another critical challenge. While full indigenisation in this area remains aspirational in the near term, diversification of supply chains and reduced exposure to single points of failure should be carefully guarded against, especially within our national security doctrines.

Conclusion

Sovereignty in India’s national security is demonstrably rising as we progressively replace rhetoric with reform. The end goal remains that India’s national security missions are planned and executed through sovereign, indigenous brains, artificial or otherwise. As we shorten the time-lapse between data collection and decision execution, India also needs to focus on a quantum-secure, reliable and independent dissemination layer. In order to achieve this degree of Atmanirbharta, we need to ensure we have end-to-end ownership or, at the very least, control over all the five layers of AI Sovereignty, deployed within our national security architecture.

(The author is a Partner at KPMG in India within the Government and Public Sector space dealing with Digital in Defence and Homeland Security.)

(The opinions expressed in this article are those of the author and do not purport to reflect the opinions or views of THE WEEK.)

TAGS