For about a fortnight in June, Washington hit the kill switch on three frontier AI models of Anthropic AI and OpenAI, keeping almost everyone in the dark outside the U.S.
The export control ban over Claude Mythos, Claude Fable and GPT-5.6 raised eyebrows from Canada to France to India.

French President Emmanuel Macron was the first one to call it out–he said some countries are using Sovereign AI as a geo-political tool, while inaugurating the Bharat Innovates event in Nice alongside Indian Prime Minister Narendra Modi.

U.K. termed it as an unprecedented crisis to its AI sovereignty while Canadian PM Mark Carney raised alarms about his country’s AI dependency.

Even as the US finally allowed Anthropic to restore Claude Fable, the kill switch move by the White House was a lesson in dependence” on intelligence built elsewhere, and capable of being switched off elsewhere.

Few put it as bluntly as Amitabh Kant.

Former G20 Sherpa and former CEO of NITI Aayog–India’s premier policy think tank, Kant minces no words as he took to X (formerly Twitter) to write, “We should stop pretending that technology is not being used as a geopolitical tool,” pointing to the ban and to GPT-5.6 being tested by the US government before anyone else could touch it.

Kant has a solution for this tech quagmire: AI must be “both sovereign and shared” — nationally controlled but universally interoperable, the way TCP/IP made the early internet.

We should stop pretending that technology is not being used as a geopolitical tool. The US Fable Ban. ChatGPT 5.6 being tested by the US government before anyone else gets access. This is the direction the world is moving in.
AI must be both sovereign and shared . The internet… pic.twitter.com/LqvxbrwhHe

— Amitabh Kant (@amitabhk87) June 29, 2026

Kant’s tweet had our attention and AI FrontPage correspondent Supriya Singh reached out to get his insights into what “sovereign and shared AI” should actually mean for India.

In an exclusive interview with AI FrontPage, Kant draws the hard line between remaining sovereign and technological isolationism, importance of utilizing shared AI resources and why India can’t afford to not become India only.

Below are excerpts from our conversation with Amitabh Kant.

Note: Kant responded in writing to AI FrontPage’s questions; his statement has been segmented for readability and published in full with his permission.

Question: You’ve argued AI must be both sovereign and shared, and you point to the internet, universal protocols and national networks, as the template. So for a country like India, which doesn’t yet have frontier models of its own, what does ‘sovereign and shared’ actually look like on the ground?

Amitabh Kant: “Sovereign and shared” should not mean choosing between building in India and accessing the world. It should mean ensuring that India has the capability, leverage and institutional confidence to bring the best intelligence available, whether developed domestically or globally, to every citizen and enterprise in ways that serve Indian priorities.

He then explains why India cannot afford technological isolation in the age where artificial intelligence is increasingly getting adopted across sectors.

“For India, sovereignty cannot sensibly mean technological isolationism. AI is being built through a deeply global innovation system. The chips may come from one geography, foundational research from another, cloud infrastructure from another, and applications, data, talent and use cases from India,” adds Kant who remained G20 Sherpa from 8 July 2022 to 16 June 2025, steering India’s 2023 presidency.

“Trying to replicate every layer behind national walls would be prohibitively expensive, slower, and ultimately deprive Indian citizens and businesses of the best tools available,” he adds.

Question: So, do we build our own frontier capability, negotiate for access to gated foreign models, or push for a shared protocol layer the way India did with digital public infrastructure?

Amitabh Kant: India first must not become India-only. The real objective should be fit-for-purpose AI. A farmer in Bihar, a teacher in a government school, a doctor in a district hospital, or a small business owner should not have to wait for an Indian frontier model to become the best in the world before they can benefit from world-class intelligence. They should have access to the strongest, safest and most affordable tools available, including models built in India and models built elsewhere.”

Kant then explains that ensuring that access would require three things happening together.

“First, India must invest seriously in domestic capability. This means compute, high-quality Indian datasets, research institutions, local language models, evaluation systems, AI safety capacity, talent and startups. India does need frontier ambition, not because every country must win the global model race, but because domestic capability gives us choice, bargaining power, resilience and the ability to build for our own languages, sectors and public needs. It also ensures that Indian researchers and entrepreneurs are not merely consumers of intelligence built elsewhere.

Second, India should remain deeply connected to the global frontier. We should negotiate reliable, competitive access to leading international models, cloud infrastructure and research partnerships. The right question is not whether a model was trained inside India or outside it. The question is whether India has meaningful access, appropriate safeguards, pricing power, portability, accountability and the ability to build value on top of it.

A country is not sovereign because it owns every technology it uses. It is sovereign when it is not dependent on a single provider, a single geography or a single political decision. That means diversified partnerships, interoperable systems, domestic capacity and the ability to switch, adapt and build.

Third, India should help create the shared layer. This may be India’s most important contribution. Just as India built digital public infrastructure that allowed many public and private actors to innovate on common rails, it can help shape open protocols for AI access, identity, consent, datasets, safety evaluations, provenance, agent interoperability and public-interest deployment,” said Kant.

Kant explains how India can create an ecosystem of shared AI resources.

“The goal is not one national model that does everything. It is an ecosystem where Indian users, startups, researchers and public institutions can safely use and combine the best models from India and the world. A small Indian startup should be able to build on an Indian language model, a global frontier model, Indian public digital infrastructure and local data, without being locked into any one company or country,” said Kant.

Lastly, vouching for the current Indian PM Narendra Modi, Kant says that the Indian government’s vision of AI for All bats for a shared ecosystem of resources so that the benefits of AI reach the last man.

“This is where the Prime Minister’s vision of AI for All becomes concrete. AI for All cannot mean AI made only by India. It must mean the best of intelligence, from within India and across the world, made available to every Indian in a way that is affordable, trusted, locally relevant and aligned with India’s development priorities,” he said.

Also Read: PM Modi Calls for Human-Centric, Inclusive AI at VivaTech 2026

“India should build. India should partner. India should set standards. India should create shared rails. And India should remain open enough to ride the full force of global innovation, while strong enough to shape where that innovation goes. That is what sovereign and shared should mean in practice: India rooted in its own capability, connected to the world, and building an AI ecosystem that serves its people first without cutting itself off from humanity’s collective intelligence,” said Kant.

Also Read: India’s Tryst with AI: White Paper Offers Roadmap for Foundation Models