At a time when artificial intelligence is fast becoming the backbone of governance, defence and economic power, a high-level session during the India AI Impact 2026 at Bharat Mandapam in New Delhi brought together policymakers, global technology leaders and industry experts to deliberate on a pressing question: Can India secure its digital future without owning the layers that power it?
The panellists suggested that digital sovereignty is about control - of data, of infrastructure, of models, and of the very cognitive systems that increasingly influence decision-making.
Brijesh Singh, IPS, Principal Secretary and DGIPR, Government of Maharashtra, brought the conversation down from theory to practice. Modern policing, he noted, is already inseparable from technology. Whether it is monitoring crowd flows at large gatherings like the Kumbh Mela or deploying number plate recognition systems across cities, AI tools are deeply embedded in operational frameworks.
But there is a catch. Models built without local context can carry error margins of up to 40 percent. An algorithm that performs well elsewhere may falter in India. "It is with all the model across the world," he said.
The real concern, Singh suggested, is dependence. If sovereign systems are essentially imported solutions, designed and controlled externally, the illusion of control can collapse at a critical moment. Sovereignty must therefore exist at every layer, data, infrastructure, governance architecture and what he described as “cognitive infrastructure.” India, he argued, has no shortage of talent. What it needs is the confidence to build and trust its own stack.
The discussion also included insights from Ajay Singhal, IPS, Director General of Police, Haryana, and Lt. Gen. Harsh Chibber, Director General of Information Systems, Indian Army, who emphasised the growing role of secure digital systems in internal security and defence operations.
If the policing lens focused on application, Martin Willcox, Global Head of Analytics at Teradata, shifted the attention to architecture. There is, he said, no good AI without good data.
For sovereign AI to work, India must modernise and integrate its data ecosystems. This means creating a platform that can cut across domains.
Willcox warned described a concept known as the “training trap”, the urge to pour enormous resources into building massive proprietary foundation models from scratch. Open-source models, he observed, are rapidly catching up with proprietary ones. Rather than obsessing over training, India might be better served by focusing on inference, deploying AI systems that are efficient, faster and more cost-effective.
In national security contexts, structured data still dominates. The answer, he suggested, is not one monolithic model but a toolbox, interoperable systems that can be inspected, adapted and trusted.
For Preet Saxena, Global Lead for Data and Analytics at Concentrix, the state cannot walk this path alone. The private sector is already building prototypes capable of scaling. Structured collaboration — from hackathons to public-private partnerships — can accelerate deployment while distributing risk.
But speed cannot come at the cost of responsibility. AI infrastructure consumes energy, shapes labour markets and influences governance. Questions of sustainability and ESG must become part of the sovereign AI conversation.
From the vantage point of global technology deployment, Mandar Kulkarni, National Security Officer (India and South Asia) at Microsoft, framed sovereignty as a layered construct. He further defined four layers: Data Sovereignty to ensure national data is not accessed or exploited beyond intended jurisdictions; Operational Sovereignty to guarantee continuity of critical services; Technology Sovereignty to build or control the technological stack that underpins digital systems; AI Sovereignty to ensure that models are trained on Indian data, reflect Indian ethics and are explainable.
AI sovereignty, he noted, is distinct from data sovereignty. It raises deeper questions: What data was the model trained on? Does it reflect Indian realities? Can its decisions be explained? The “black box” problem becomes more than a technical issue; it becomes a strategic one. The "black box problem" refers to the inability to understand, interpret, or trace how complex AI systems, particularly deep learning and neural networks, arrive at specific decisions or predictions.
Pier Stefano Sailer, Global Lead for Digital Sovereignty at KPMG, argued that countries often swing between two extremes — total state control or unrestrained private dominance. India, he suggested, has the opportunity to chart a middle path, combining strong public digital infrastructure with private innovation and investment.
By the end of the session, one theme stood out. Digital sovereignty is not about isolation. Nor is it about rejecting global collaboration. It is about ensuring that the foundational layers of AI, the data, the compute, the models and the governance frameworks, are aligned with national priorities.
