The Third Option

I. The Framework and Its Assumptions

The most rigorous recent attempt to map the strategic terrain of advanced AI governance comes from the Institute for AI Policy and Strategy, whose work on navigating the transition to transformative AI posits a foundational problem that conventional diplomacy is ill-equipped to solve.

The core challenge — what IAPS terms the decisive strategic advantage (DSA) commitment problem — is that any actor racing toward superintelligence faces structural incentives to defect from cooperative arrangements the moment it believes it can win unilaterally. Traditional treaty architecture, built on verified compliance and mutual deterrence, presupposes actors who are already in equilibrium. Advanced AI disrupts that equilibrium before the treaties can be written.

From this starting point, IAPS derives nine distinct strategic visions for how the transition might be navigated. Some locate decisive authority in a dominant United States, others in a stabilised mutual assured AI malfunction (MAIM) equilibrium between Washington and Beijing,[1] and others still in a global regulatory body with genuine enforcement power. Each vision is internally coherent. Each offers a plausible account of how catastrophe might be averted. And each shares a common architecture: the world is divided between two primary actors whose choices determine outcomes, and every other nation is either an allied asset — a supplier of compute, a host for data centres, a signatory to US-drafted safety frameworks — or a passive recipient of what IAPS calls "economic superabundance," the flood of productivity gains that a successfully governed AI transition is expected to deliver to all.[2]

This is a framework of considerable analytical power. It is also, when tested against the events of February 2026, a framework that reaches the edge of its current map. IAPS's nine visions account, between them, for every plausible configuration of Washington and Beijing. None of them adequately account for New Delhi.

II. What India Just Did

In February 2026, India hosted the AI Impact Summit in New Delhi, convening more than one hundred nations and over twenty heads of state. The attendee list read less like a diplomatic courtesy call and more like a restructuring of the field's centre of gravity: Sam Altman of OpenAI, Dario Amodei of Anthropic, Demis Hassabis of Google DeepMind, and Sundar Pichai of Google were all present.[3] For a single convening, in a country that Western AI governance discourse has tended to treat as a downstream recipient rather than an upstream architect, this was a remarkable feat of agenda-setting.

IAPS's own analytical memo on the summit notes something important — a memo that itself flags whether India can move beyond aspirational framing as the defining question going forward:[4] the New Delhi convening shifted the field's emphasis from frontier AI safety, the preoccupation of the Bletchley and Seoul processes, toward "demonstrable impact," toward applied AI's immediate consequences for healthcare access, agricultural productivity, and public service delivery. This was not a rhetorical accident. It reflected a deliberate choice about what an AI governance forum looks like when it is designed by and for a nation of 1.4 billion people, the majority of whom will not interact with frontier models in the near term but will be profoundly affected by how those models reshape labour markets, state capacity, and geopolitical alignment.

What made the summit strategically legible was also what made it analytically inconvenient: India did not choose sides. The United States arrived with an agenda framed around domination — AI as a geopolitical race against China, its senior officials declaring that they "totally reject global governance of AI." France's Macron offered a substantively different doctrine: algorithmic transparency, child protection from unregulated digital environments, and linguistic diversity as the organising principles of AI governance — none of which map onto either Washington's or Beijing's frameworks. India hosted both delegations, absorbed neither position, and demonstrated its non-alignment in operational terms: it is simultaneously building on US foundation models and deploying DeepSeek, maintaining technological optionality that mirrors its diplomatic posture exactly. As Cameron Kerry of Brookings observed, India has concluded trade agreements with both the United States and the EU while continuing to trade significantly with China and Russia — a pattern of behaviour, not a sequence of coincidences.[5]

III. The Gap This Reveals

The bilateral architecture India is constructing around the summit gives the third-path claim institutional substance rather than rhetorical assertion. France and India elevated their partnership to a Special Global Strategic Partnership centred on AI, with both governments explicitly framing the relationship as a stabilising force in a period of geopolitical turbulence.[6] The EU-India Free Trade Agreement — the first comprehensive trade deal the EU has concluded with a major emerging economy in over a decade — embeds AI and semiconductor cooperation provisions within a binding multilateral framework of twenty-seven states that is neither US-drafted nor Chinese-designed. Von der Leyen framed the agreement as evidence that rules-based cooperation still delivers.[7] Macron was more direct: France, he said, is "determined to shape the rules of the game with allies such as India."[8]

In IAPS's framework, this pattern looks like ambiguity. Examined more carefully, it is a strategic position — and one now backed by treaty architecture with twenty-seven states simultaneously. That position, when stress tested against IAPS's own framework, reveals two specific gaps, each pointing to an underexplored dimension of the global order.

The first gap concerns the benefit-sharing model embedded in every IAPS vision that addresses developing nations. The underlying assumption is that nations outside the US-China frontier will require managed access to AI's gains — subsidised compute, transferred models, multilateral redistribution mechanisms. India's sovereign stack construction challenges this assumption at its foundation. Under ISM 2.0, India is crowding private investment into the chip layer through semiconductor policy architecture. Budget 2026's tax holiday for foreign cloud providers is made conditional on the use of Indian data centres and local resellers, an incentive structure that secures the storage and infrastructure layers without prohibition. At the model layer, indigenous large language models are already assessed as capable of solving ninety-five percent of domestic use cases.[9] This five-layer construction is a rational structural response to a demonstrated vulnerability: export controls on advanced computing chips have shown that AI superpowers can throttle other countries' ambitions by controlling supply chain chokepoints. India builds because structural dependence on compute that rival powers can withdraw through export controls is not a risk any serious state actor can absorb.[10]

That said, intellectual honesty requires the qualification that India's own UPSC Chairperson Ajay Kumar has acknowledged publicly: GPU capacity remains critically low compared to the United States and China, and "computing without access, data without governance, and models without accountability will only deepen dependence — not power."[11] India is a serious actor constructing a sovereign stack under real and significant constraints. A serious policy analysis must hold both facts simultaneously. The category of "bystander to be stabilised with handouts" does not fit this nation, and the IAPS framework's strategic visions have not yet mapped a category for builders who are neither American nor Chinese — India is evidence that this category exists and matters.

The second gap is more consequential for the specific crux that IAPS itself identifies as the hardest unresolved problem in AI governance: how do you construct international cooperation frameworks when the relevant actors are strategic rivals with deep mutual mistrust?[12] IAPS acknowledges this openly. Its most hopeful visions depend on trust-building mechanisms, verification regimes, and diplomatic bridging capacity that are described as necessary but left analytically underdeveloped. India, it turns out, is precisely the actor that such mechanisms require. It purchases Nvidia chips and maintains military and energy ties with Russia. It hosts American chief executives and declines to sign American-drafted declarations. It deepened its AI cooperation with US-aligned Israel — signing an MoU on artificial intelligence, establishing a Centre of Excellence in Cybersecurity, and launching a Critical and Emerging Technologies Partnership — and simultaneously declined to be absorbed into Netanyahu's proposed Hexagon alliance, even as Netanyahu publicly named India within it. That is not passive hedging. That is active non-alignment operating under direct and named pressure.[13]

This is not incoherence. This is the profile of an actor with credibility on both sides of the central rivalry — and credibility of precisely this kind is what makes verification and trust-building between rivals possible at all. India's strategic non-alignment, which IAPS's framework has no category for currently, is not a governance liability. It may be the most valuable governance asset currently available to the international system.

IV. What India Must Do Now

Recognising this position is one thing. Converting it into durable institutional architecture is another, and the window for doing so is narrower than it appears. Three specific steps follow from the analysis.

First, India must treat the BRICS Summit later this year as a structured AI governance moment rather than a diplomatic formality. The convening authority demonstrated in New Delhi — 100-plus nations, heads of state, the full spectrum of frontier AI leadership — is a depreciating asset. It commands attention now because it was surprising. To sustain influence, India must convert that convening authority into institution-building. A BRICS AI governance dialogue, with defined working groups, published research outputs, and a standing secretariat, would constitute the first multilateral AI governance body anchored in the Global South. That matters not only symbolically but structurally: it would give the hundred-plus nations who attended New Delhi a forum to develop and express collective positions, rather than choosing between existing Western and Chinese frameworks by default.

Second, India should propose a non-aligned AI verification mechanism — and the urgency of this prescription is sharpened, not weakened, by the Declaration's own limitations. Critics have noted that the New Delhi Declaration is voluntary and non-binding, its text marked throughout by the diplomatic formula "take note of" rather than meaningful endorsement, with no body empowered to ensure that signatory nations follow its principles.[14] This makes the Declaration not a solution to the verification problem IAPS identifies, but an illustration of it: another charter added to a proliferating list of initiatives without enforcement mechanisms, spreading political will thin without allowing anything to take root. The contrast with the Seoul Frontier AI Safety Commitments — which required signatories to assess risks and define intolerable thresholds — is pointed.[15]

India's prescription should therefore be precise: build the verification body that its own Declaration conspicuously lacks. The institutional precedent exists — India's historic role in the Non-Aligned Movement's observer mechanisms offers a template, however imperfect, for what positional independence in verification could look like in practice. Neither the United States nor China can play the verification role relative to the other. India, with appropriate institutional design and genuine political commitment, could.

A sceptical reader might object that India's governance independence is structurally fragile — that Washington can always impose costs on Delhi for non-alignment through trade coercion. The United States Supreme Court's recent 6-3 ruling that the International Emergency Economic Powers Act does not authorise sweeping open-ended presidential tariffs substantially constrains that tool.[16] India's continued purchase of Russian oil — the most visible test of Washington's tolerance for non-alignment — is structurally protected by precisely this constraint.[17] India's third position is more structurally durable after February 2026 than it was before it.

Third, the New Delhi Declaration's silence on frontier AI safety was a missed opportunity that must not be repeated. The agenda India set in February was valuable, but agenda-setting is not agenda-ownership. Singapore and Switzerland will host subsequent convenings in the established governance sequence,[18] and without a prior Indian articulation of how frontier AI safety and development impact can be made conceptually coherent — rather than treated as competing priorities — the field will revert to Western framings by default. India has perhaps six months to develop and publish a substantive position on frontier AI governance that reflects its unique vantage point. Failing to do so would mean that the most consequential question — what governance norms apply to the most powerful AI systems — gets answered in rooms where India is a guest rather than a host.

Conclusion

IAPS's strategic visions are rigorous, analytically serious, and important. They should be read carefully by anyone who hopes to reason clearly about the governance of transformative AI. But they were written for a bipolar world, and the 2026 AI Impact Summit demonstrated that the most consequential AI governance decisions may not ultimately be made in Washington or Beijing.

They may be made in the hundred-plus nations who came to New Delhi — nations that are right now choosing their technical stack, their strategic partners, and their governance models, in most cases without any adequate framework to guide those choices. India's decision to host that summit was significant. What it does with the authority that hosting conferred will determine whether it becomes a genuine architect of international AI governance, or merely a well-attended venue on the way to somewhere else.