The Trump administration's June 2026 order compelling Anthropic to suspend access to its Fable 5 and Mythos 5 models for every foreign national — inside or outside the United States, including the company's own overseas staff — is not a distant American quarrel that the Global South can afford to ignore. For the first time, Washington has extended export controls from the hardware of artificial intelligence to the models themselves, treating a frontier system as a strategic asset to be switched off at will. The lesson is uncomfortable but unambiguous: capability granted from abroad can be withdrawn from abroad.

What unsettles partners and allies alike is the logic on display. The instinct to gate technology by nationality, to police data flows, to ration chips and critical minerals, and to fold all of this into a single national-security calculus is the very instrument the United States honed in its long contest with China. That same instrument is now turned outward, indiscriminately, against friends as readily as rivals. When the criterion for exclusion is simply "foreign national", every country outside the American perimeter becomes, in principle, a candidate for the kill switch. Europe has already begun to speak of treating AI as it once treated nuclear power — a question of sovereignty rather than commerce. The Global South must reach the same conclusion, and faster.

For India, the stakes are sharpest at the level of compute. New Delhi has made genuine progress: the IndiaAI Mission, backed by some $1.25 billion, has underwritten indigenous foundational models, and Sarvam AI's 30-billion and 105-billion-parameter systems were trained from first principles on domestic infrastructure across twenty-two Indian languages. Yet that infrastructure still runs almost entirely on imported silicon. A model conceived, trained and deployed in India remains tethered to a supply chain over which India exercises little control. Legislation now moving through the US Congress would prioritise American customers for advanced chips before any are exported — a reminder that the binding constraint is not the algorithm but the wafer.

This is where dependence becomes dangerous. If India — or any aspirant in the Global South — remains reliant on the United States, or indeed on any single power, for chips and semiconductors in the years ahead, that dependence will function as a mortgage on sovereignty. The collateral is not merely commercial access but the autonomy to govern one's own data, defend one's own infrastructure, and choose the terms of one's own cooperation. Export leverage rarely stays confined to its original domain; a power willing to withhold AI today can condition trade, defence or technology partnerships tomorrow.

The answer is not autarky, which is neither feasible nor desirable. It is strategic autonomy built deliberately across the stack. India must consolidate and scale what Sarvam represents — not as a showcase but as a foundation — while pressing on the harder frontiers of domestic fabrication, advanced packaging, and a chip supply diversified through partnerships that are plural rather than singular. Data centres on Indian soil, governed by Indian law, are a necessary hedge against the externalisation of data sovereignty that the model ban foreshadows. So too is South–South coordination: shared compute, common standards and pooled bargaining power can convert a collection of dependent consumers into a constituency that shapes the technology rather than merely receiving it.

The Anthropic episode will pass; the precedent will not. The Global South should treat it as the clarifying moment it is — proof that in the age of intelligent machines, the line between a technology partner and a technology gatekeeper is drawn in Washington, not in New Delhi. Sovereignty now begins with the capacity to keep the lights on for oneself.