The AI arms race is moving from models to national security infrastructure

By Tarun Wig, Co-founder and CEO, Innefu Labs Limited.

For the past few years, most of the conversation around artificial intelligence has come down to one question: whose model is bigger, faster, and smarter? Every few months there’s a new benchmark, a new billion-dollar training run, a new claim to being at the frontier. It makes for good headlines. But I think we’re watching the wrong contest.

The more consequential race is happening at a level below the models themselves. It is a race over who controls the infrastructure that keeps nations safe in an age when AI can be used to attack them just as easily as it can be used to defend them. That shift, from “who has the smartest model” to “who controls the systems protecting critical infrastructure and citizens”, is where the real strategic stakes now sit.

Consider how the nature of threats to a country has changed in a fairly short space of time. Financial fraud used to mean a forged signature or a cloned card. Today, it can mean a synthetic voice on a phone call, indistinguishable from a senior bank official, authorising a transaction that never should have happened. In the last year, Indian financial institutions have already seen a dramatic uptick in just such an attack. AI can be used to trick companies that were never designed to ask if the caller on the other end was a real person. Fraudsters are using AI voice, face and identity to beat systems that were not created to ask whether or not the person on the other end is a real person.

Organised crime has evolved as quickly. Criminals have turned to AI tools to create increasingly long and difficult-to-trace money laundering pathways, produce large volumes of falsified documents, and orchestrate multi-border crimes more quickly than traditional investigations can match. Then there’s the information layer. Deepfakes and synthetic media aren’t a novelty anymore; they’re some entertaining tricks confined to social media. They are being used to influence public opinion, to impersonate public figures, and to muddy the waters in high-stakes situations, such as elections, communal violence and the timing of market-sensitive announcements. A recent McAfee survey shows that almost half of Indian adults have suffered, or know someone who has suffered, from an AI voice cloning or deepfake scam; that’s almost twice the global rate.

Detecting all of this requires AI systems of real sophistication, ones that can spot the statistical fingerprints of synthetic content, flag anomalous financial behaviour before it becomes a loss, and connect disparate signals into something an analyst can act on. Building and running that kind of capability is now as much a part of national security as border patrols or radar systems ever were.

Which brings up an uncomfortable question that not enough people in India are asking out loud: who is actually building and operating these systems, and where is the sensitive data flowing?

It is entirely reasonable for a country to use the best available technology, wherever it comes from. But there is a meaningful difference between using a foreign AI platform for a general productivity task and depending on one to analyse intelligence, screen financial crime, or monitor threats to critical infrastructure. When the underlying models, the training data pipelines and the operational control of these systems sit outside the country, you end up with a dependency that’s very hard to undo later. Access, updates or support can become geopolitical bargaining chips in certain situations, especially in a crisis.

This is where sovereignty gets misunderstood. It doesn’t mean India has to build every layer of the AI stack from raw silicon upward. That’s neither realistic nor necessary, and chasing it just wastes energy on the wrong fight. What sovereignty actually needs is the capability, the talent and the technological control to deploy AI on India’s own priorities and India’s own terms, answerable to Indian institutions, not a foreign vendor’s roadmap.

There are early signs India is taking this seriously. The IndiaAI Mission, backed by an outlay of over 10,000 crore rupees, is deliberately structured around building indigenous compute capacity, domestic models and homegrown talent rather than simply buying capability off the shelf. That is the right instinct. The next step is extending it deliberately into the domains where sovereignty matters most: intelligence, law enforcement, defence and the operators of critical infrastructure.

There is also a bigger opportunity sitting inside this, beyond just defending our own systems. India already has a strong base of deep-tech talent working on computer vision, natural language processing and analytics for security use cases. Give that talent the mandate, the data access and the sustained investment to build serious, indigenous security AI products, and there is no reason those products need to stay within our borders. Technology that holds up under the demands of Indian government deployments, some of the toughest operating conditions anywhere, tends to hold up well elsewhere too. Countries across the Global South, many of whom share India’s caution about depending entirely on Western or Chinese security platforms, are looking for credible alternatives.

The AI race the world is fixated on, over model size and benchmark scores, will keep making headlines. But the quieter race, over who builds and controls the infrastructure that keeps a nation’s data, institutions and citizens safe, is the one that will matter more in a decade’s time. India has a narrow but real window to lead in that race rather than simply consume what others have built. Sovereignty, capability and commercial opportunity are, for once, pointing in the same direction. It would be a shame to look back and realise we chose to watch someone else’s race instead of running our own.