The West’s advantage in artificial intelligence depends on an increasingly integrated ecosystem of private companies, governments, intelligence partnerships and military alliances. Yet the most advanced capabilities underpinning that advantage remain subject to national control, above all that of the United States. As AI moves rapidly into security and defence policy, the gap between the collective nature of Western interests and sovereign control over the technologies needed to protect them is acquiring strategic significance far beyond the familiar debate over the relationship between governments and Big Tech.
Last month’s G7 summit in France offered a particularly vivid illustration. The leaders of the world’s largest AI companies sat alongside heads of state and government to discuss technology, security and the future of Western power. OpenAI’s Sam Altman held bilateral meetings with G7 leaders and other invited heads of government, following an agenda and protocol normally associated with state actors. Others, including Anthropic’s Dario Amodei and Google DeepMind’s Demis Hassabis, were central figures in discussions about the technological leadership of democratic countries.
Beneath the image of a Western bloc seeking to preserve its technological superiority lies a more complex reality. A significant share of the most advanced models, computing infrastructure and talent required to develop them is concentrated in private companies. States retain political authority, regulatory power and control over military force, yet depend on technological capabilities they often neither develop nor own. For alliances, this dependence creates an additional problem: integrating technologies produced by private actors and subject to national jurisdictions into collective security structures.
The extent to which this interdependence remains exposed to the decisions of individual governments had become clear even before the summit began. While leading AI executives urged Western democracies to avoid fragmentation and to develop structured access to frontier models, the US government imposed restrictions on non-US users’ access to Fable 5 and Mythos 5, including those in allied countries. The controls were later lifted and replaced with temporary access measures, amid formal criticism from the European Union of the Trump administration’s export control policies. The episode does not necessarily point to a lasting rupture between Washington and its allies. It does, however, show how easily tensions can emerge between two trends that are advancing in parallel.
The AI industry operates across borders, and emerging security architectures require ever deeper technological integration. However, control over the most advanced capabilities remains national and can be exercised unilaterally when governments judge their security interests to be at stake. For US allies, the question therefore concerns the predictability of access to technologies on which an increasing share of collective security may depend. This would matter even under conditions of stable technological leadership. It becomes more consequential when the durability of that advantage is difficult to measure and the time available to translate it into strategic capability appears to be shrinking. Ten days after the order affecting Anthropic’s models, the cyber security chiefs of the Five Eyes intelligence partnership warned that the rapid development of frontier models could render existing cyber risk assessments obsolete “in months, not years”. Their concern was the growing capability available to adversaries.
The same uncertainty is evident in assessments of the distance between US and Chinese models. The Center for AI Standards and Innovation (CAISI) has estimated that DeepSeek V4 Pro trails leading American frontier models by about eight months, based on its own benchmark suite. Results published by DeepSeek, using different tests, instead place the model close to US systems released roughly two months earlier. The divergence does not establish how quickly China is closing the gap, but it does suggest something equally significant: the scale and durability of America’s technological lead have become difficult to capture in any single, authoritative measure.
Nor is the pressure created by China’s progress confined to competition between models. Beijing is accelerating the integration of artificial intelligence into military capabilities, spanning logistics, decision support, and autonomous systems. The challenge for the West therefore has two dimensions that increasingly reinforce one another: technological advantage must be preserved while governments and alliances seek to convert it into operational capability. Yet the time required for the second process may exceed the duration of the first.
It is precisely this gap that gives the NATO summit of 7-8 July its significance. The Alliance, which often serves as a reference point for operational standards and strategic priorities beyond the strictly military sphere, announced plans to develop an “interoperable transatlantic warfighting cloud” and to adopt “powerful artificial intelligence models” as part of the modernisation of its military capabilities. The project aims to create an infrastructure through which data, sensors, AI models and autonomous systems can operate across the armed forces of different countries, integrating technologies developed by different companies and subject to different national systems into a common military architecture.
In such a system, interoperability ceases to be an exclusively technical concern. An integrated military architecture requires sufficiently predictable access to capabilities that allied governments may neither own nor ultimately control. The more Western armed forces depend on advanced models and private infrastructure, the greater the risk that national decisions on access will affect the system as a whole. The problem exposed by the US restrictions thus acquires an operational dimension: the integration of military capabilities may advance faster than the political arrangements governing the technologies that make such integration possible.
NATO was already confronting the difficulty of adapting its institutional timelines to the pace of technological change. The Rapid Adoption Action Plan, approved at the 2025 Hague summit, aims to integrate new products into allied armed forces within a maximum of 24 months. Reducing adoption cycles to two years is an ambitious objective for an alliance of 32 countries. The relevant timescale changes, however, when intelligence agencies warn that the AI threat landscape may evolve within months and estimates of the distance between American and Chinese models are increasingly expressed in the same unit of time.
Artificial intelligence poses a less familiar challenge for military alliances. Conventional weapons systems are acquired, maintained and upgraded through industrial dependencies that have developed over decades. Advanced models, by contrast, may require continuous access to computing infrastructure, updates, data and the capabilities of their providers. Dependence, therefore, does not end with the acquisition of the technology. It may persist throughout its operational lifecycle and leave the alliance exposed to changes in political relations, trade rules and the decisions of the country exercising jurisdiction over the provider.
The evolution of drones and autonomous systems makes the consequences of this shift more tangible. The military value of these technologies increasingly depends on the ability to connect sensors, data, models, cloud infrastructure and decision-making processes within continuously updated networks. This is the logic behind NATO’s proposed warfighting cloud, and it is also the reason access can no longer be separated from interoperability. Integrating military systems, digital infrastructure and privately developed technologies within the same operational environment creates new capabilities, but also new forms of dependence.
From this perspective, the images from the G7 take on a more complex meaning. AI executives now sit at the table with governments because they control capabilities that are now relevant to both national and collective security. Their presence reflects the degree of interdependence within the Western ecosystem without diminishing the enduring weight of sovereignty. When security interests diverge, governments retain the power to determine who can access technologies developed within their jurisdiction.
For the United States and its allies, the challenge is to manage requirements that do not necessarily align: ensuring sufficiently stable access to critical AI capabilities to sustain Western strategic integration while preserving national control over technologies Washington considers central to its security. Agreements governing access to models, common standards, guarantees of operational continuity, and clearer arrangements between private companies and alliances may therefore become an essential component of the Western security architecture.
Otherwise, technological integration risks proceeding on the basis of an assumption that recent months have already called into question: that the availability of the most advanced capabilities within the Western ecosystem means allies can rely on stable access to them. The gap between those two conditions may prove to be one of the most difficult strategic problems that artificial intelligence poses for Western alliances.
The copyright of this article remains with the Politeia Research Foundation ©️PRF.
