Anthropic blackout: This is what strategic AI dependence looks like

For years, concerns about technological dependence were brushed aside as protectionist anxiety.

6 min readJun 19, 2026 02:39 PM IST

First published on: Jun 19, 2026 at 02:39 PM IST

The US government’s decision to abruptly suspend foreign access to Anthropic’s most advanced AI models, Mythos 5 and Fable 5, may prove to be one of the defining technology policy moments of this decade.

Just two weeks earlier, India had reportedly secured access to Mythos-class models through Project Glasswing, Anthropic’s cybersecurity initiative, after sustained lobbying. While details of the exact model that was made available are not public, this access was meant to be narrow and specific, intended only for defensive protection of critical digital infrastructure. The EU’s cybersecurity agency ENISA and other institutional partners across a host of countries were also granted access. But none of that mattered. What the Friday order shows is that even carefully negotiated and tightly restricted access can be wiped out overnight.

This begs a question: If technology cannot even be shared with countries that are close allies, what are those relationships actually worth, and what would happen if geopolitical interests were actually to diverge?

For years, concerns about technological dependence were brushed aside as protectionist anxiety. The assumption was that in an interconnected world, access to frontier technology would flow through open markets. Unfortunately, that assumption has now come undone, and the consequences will extend well beyond geopolitics.

The episode also carries a cautionary lesson for American tech companies. A firm that complies with the law at home risks becoming unreliable abroad. American tech has spent decades convincing governments and businesses around the world to trust their infrastructure and build core operations on their services. That trust is not going to be easy to rebuild once broken. When customers start believing that access can be switched off by executive order, they are bound to start making different procurement decisions. First quietly and slowly, and then all at once.

For India, the lesson is fairly clear. A country that is heavily dependent on foreign technology players for critical digital services is carrying a structural vulnerability that no diplomatic arrangement can fully cover.

And that vulnerability is especially relevant in the context of the current debate on India’s technological pathway. One side of the debate, articulated most prominently by Nandan Nilekani, argues that India’s comparative advantage lies not in building another frontier model but in becoming the world’s leading use-case and application economy. The argument is certainly compelling from a productivity point of view. But Friday’s developments expose its limits: What may be economically efficient is not always geopolitically sufficient.

The answer, however, is not to swing to the opposite extreme and build everything domestically. The AI value chain, from chips to data centres to foundational models to applications, is deeply global, and no country can realistically delink from it entirely. The vast majority of AI use cases carry no meaningful strategic risk either. The right response, therefore, is not a panic-driven push for self-sufficiency across the entire value chain but a more pointed question: Which dependencies actually create strategic exposure, and which parts of the stack can safely remain in the hands of global providers? We have barely begun to engage with this question, and certainly not at the level of rigour it demands.

The India AI Mission and the Ministry of Electronics and Information Technology’s (MeitY) ongoing work on sovereign cloud are steps in the right direction and reflect a recognition that strategic capability cannot be left entirely to market forces. But we need to be more comprehensive in our approach, and more operationally precise about how we define and operationalise sovereignty. Sovereign tech cannot mean the same thing for defence as for a government ministry using AI for grievance redressal, or for a startup generating product descriptions. A few questions need to be answered before India can claim to have a sovereignty framework worth the name: Which parts of the value chain — and for which sectors — must sit under Indian control? Is it about where data is processed, who controls the model, or who owns the infrastructure where the model sits?

Getting this right is important because there are real trade-offs at stake. Indian businesses need access to the best available technology at affordable prices to remain globally competitive. Imposing additional constraints when there is no strategic imperative will drive up costs without improving security. We also need to acknowledge that India’s R&D and capital resources are limited (we spend less than 0.7 per cent of our GDP on R&D, compared with more than 2.5 per cent in China and over 3 per cent in the US), and therefore our spending needs to be concentrated rather than spread thin.

Identifying areas that are critical for strategic independence and investing deeply in them will be far more effective than trying to do everything at once. Top of the list would be compute infrastructure for sensitive workloads, cyber defence capacities, and targeted frontier AI capabilities, where denial of access would affect India’s resilience. For most else, India’s focus should remain on rapid adoption and access to the best technology available globally.

Last Friday’s blackout will eventually be resolved. But the episode is a clear demonstration of what strategic AI dependence actually looks like in practice. The question now is whether we treat this as a wake-up call or keep kicking the can down the road until someone else pulls the plug.

The writer is the founding partner at the public policy firm The Quantum Hub (TQH)