Sovereign AI: the G7 worries about an American “kill switch”

At the G7, dependence on American AI becomes a matter of political security

The debate over artificial intelligence has long been dominated by the race for models, chips, and use cases. But at the G7 summit, another angle forcefully emerged: strategic dependence. As TechCrunch reports in its article “World leaders want American AI. They just don’t want America to be able to turn it off.”, several leaders expressed increasingly explicit concern about the central role of American providers in global AI, and above all about the possibility that access to these services could be interrupted for political reasons.

At the heart of the problem is a phrase that has become especially evocative: the “kill switch”. The expression does not necessarily refer to a single button or a standardized technical function. More broadly, it refers to the control capability retained by a provider, or by the state on which it depends, over critical AI services accessed remotely. In a market where a large share of advanced capabilities relies on APIs, cloud, and infrastructure operated by a handful of American groups, the question is no longer only who innovates the fastest. It becomes: who can cut off access, under what conditions, and with what consequences?

According to TechCrunch, this concern was voiced at the highest political level, notably by Emmanuel Macron and Narendra Modi. The signal is important, because it shows that the discussion around sovereign AI is no longer confined to technical circles, European regulators, or advocates of digital autonomy. It is moving to the level of heads of state, in a framework where AI is now seen as strategic infrastructure, on the same level as energy, telecommunications, or semiconductors.

This shift is not happening by chance. Since the rise of generative AI, the global ecosystem has been structured around a few American players that possess both the most visible models, the most powerful cloud platforms, and the largest computing capacity. For a government, a large company, or a public administration, adopting these building blocks often means gaining performance and speed of deployment. But it also means accepting a form of contractual, technical, and geopolitical dependence.

The G7 acts here as a revealer. Western leaders want to benefit from American advances in AI, since they have become difficult to bypass in the short term. However, they do not want that dependence to be turned into a lever of constraint. The nuance is decisive: the issue is not a rejection of American AI, but the fear that a technology that has become essential to the economy, public administration, and potentially national security could be subject to an external power to interrupt it.

For Europe, this debate echoes an older history. Discussions about digital sovereignty did not begin with generative AI. They have run through cloud, data hosting, software dependencies, platforms, and extraterritorial rules. But AI adds a new layer, because it combines several dependencies at once:

• dependence on the models themselves;
• dependence on the APIs that make it possible to consume them;
• dependence on the hyperscale clouds that host them;
• dependence on chips and computing capacity;
• dependence on updates, filters, usage policies, and compliance mechanisms imposed by providers.

What the TechCrunch article highlights, then, is less an abstract fear than a very concrete realization: if AI becomes a critical service, then control over its availability becomes an attribute of sovereignty. And if that availability depends on foreign actors, sovereignty mechanically becomes incomplete.

What Macron and Modi are saying, and why the political signal is strong

According to TechCrunch, Emmanuel Macron and Narendra Modi explicitly highlighted the risk that access to critical AI services could be cut off by American political decision. This wording is important because it shifts the debate. Until now, most public discussions around AI focused on model safety, bias, disinformation, copyright, or the impact on employment. Here, the angle is different: it is about the continuity of access to the technology.

In other words, the question is not only: “Can these models be trusted?” It also becomes: “Can we guarantee that they will remain available if the diplomatic, commercial, or regulatory context becomes tense?”

The fact that this issue is being raised by top-level leaders changes the scale of the debate. Emmanuel Macron has for several years aligned himself with a position favorable to greater European technological autonomy, whether in cloud, semiconductors, or more recently AI. Narendra Modi, for his part, represents a country that is also seeking to strengthen its national digital capabilities while navigating a technological environment dominated by a few powers.

The convergence of these positions is revealing. It shows that the concern is not limited to the European Union or to a specifically French sensitivity around “sovereignty.” It affects states with very different economic and geopolitical profiles, but facing the same reality: cutting-edge AI is today largely supplied by American companies, often through centralized services, and that centralization creates a point of vulnerability.

TechCrunch clearly underscores this tension: leaders want American AI because it is often the most advanced, the most available, and the most integrated into digital value chains. But they do not want the United States to be able to “turn it off.” This apparent opposition sums up a structuring contradiction of the current market. The technological attractiveness of American solutions is strong, but the more their adoption spreads, the more the strategic cost of dependence rises.

The term “kill switch” also has symbolic reach. It evokes the idea of ultimate control, of an asymmetry of power between the one who provides the capability and the one who uses it. In traditional digital infrastructure, this kind of asymmetry was already familiar: account suspensions, export restrictions, limits on access to certain technologies, application of sanctions, unilateral changes to terms of use. With AI, this asymmetry becomes potentially even more sensitive, because models can be integrated into critical processes: customer relations, software production, document analysis, decision support, administrative automation, cybersecurity, or research.

For European companies, the warning is far from theoretical. Many are already deploying AI services through foreign platforms, with architectures that rely on API calls and managed environments. As long as access is smooth, the model seems rational: low friction, speed of integration, limited initial costs compared with in-house development. But if service availability becomes a political risk, then the calculation changes. The question is no longer only the price or the quality of the model. It also becomes a question of resilience.

In this context, political speech matters more than usual. When a head of state publicly mentions the risk of an AI service being interrupted by a foreign power, it sends a message to several audiences at once:

• to public administrations, which must integrate this risk into their architecture choices;
• to strategic companies, which must think differently about their dependencies;
• to European providers, which see in it a political and commercial opportunity;
• to American partners, who are reminded that trust cannot rest solely on technical performance.

This political signal could have lasting effects, even without any immediate measure. In technology markets, risk perception plays a major role. A concern expressed at the G7 level may be enough to accelerate internal audits, justify diversification budgets, or revive sovereignty projects previously deemed too costly or too slow.

The “kill switch” goes beyond models: cloud, APIs, compute, and the dependency chain

One of the most useful contributions of the debate highlighted by TechCrunch is the reminder that sovereignty in AI is not limited to the nationality of a model. The risk mentioned by leaders concerns a set of technical layers that, taken together, determine the real ability of a country or a company to use AI without excessive dependence.

The first layer is that of the models. When an organization uses a large language model or an AI service provided by an American player, it depends on that service remaining online, on its versions, its usage limits, and its commercial terms. Even if performance is excellent, control over the roadmap remains external.

The second layer is that of the APIs. They are often how AI enters business applications. An API can be modified, restricted, priced differently, or made unavailable in certain areas or for certain uses. This dependence is sometimes even stronger than dependence on the model itself, because the company’s entire software integration relies on it.

The third layer is that of the cloud. Even when a model can theoretically be deployed more autonomously, its operation at scale often depends on powerful cloud infrastructure. Yet the global cloud market is dominated by American groups. The debate over the “kill switch” therefore connects very directly with older discussions about hosting, extraterritoriality, and operational control over digital services.

The fourth layer is that of compute. Modern AI requires considerable hardware resources, especially for training and large-scale inference. The concentration of these capabilities in the hands of a small number of players reinforces dependence. Even with a good model, without stable access to compute, sovereignty remains theoretical.

Finally, the layer of software governance must be added: updates, security filters, moderation, logging, compliance, identity management, supervision. In AI as a service, these elements are essential. They are not always visible, but they condition the real use of the technology.

The central point is that the ability to “cut off” a service does not necessarily require a dramatic decision. It can take much more ordinary forms:

• restriction of access to certain features;
• inability to deploy in a given sector;
• change in compliance policy;
• geographic blocking;
• contract termination;
• prioritization of other markets in the event of capacity strain.

This reality is what gives the debate geopolitical significance. AI is not just software installed once and for all on a local machine. In its current dominant form, it is a living service, dependent on a chain of infrastructure and governance. Sovereignty must therefore be thought through across that entire chain.

For Europe, this refers back to already familiar debates around trusted cloud, data localization, and strategic autonomy. But AI makes the issue more urgent. Where a traditional software dependency could sometimes be worked around gradually, dependence on AI services integrated into operations can become much harder to replace quickly. The more an organization automates, the more sensitive it becomes to an interruption.

This reality also explains why the issue interests governments as much as large companies. In sectors such as finance, healthcare, industry, energy, or public services, AI is expected to be inserted into essential processes. If those processes rely on an external building block whose availability is not fully controlled, the risk no longer falls only under the CIO. It becomes strategic.

The notion of a “kill switch” therefore serves as a political shorthand to describe systemic dependence. It does not refer only to an extreme scenario of abrupt shutdown. It designates the full set of control levers that a provider or its legal environment can exercise over a technology that has become critical. It is precisely this shift in perspective, from performance to control, that marks a turning point in the international discussion on AI.

Why this concern could accelerate sovereign AI in Europe

The debate brought up by TechCrunch could have an accelerating effect on European policies around local alternatives, multi-vendor, and sovereign AI. The issue is not new, but it changes status when framed in terms of security of access rather than industrial preference. It is one thing to support local champions in the name of competitiveness; it is another to do so in the name of operational continuity and strategic autonomy.

In Europe, this shift is important. Debates on digital sovereignty have sometimes been seen as abstract, costly, or defensive in the face of the American lead. The “kill switch” argument gives them a more concrete dimension: it is no longer only about having European alternatives in principle, but about reducing a risk of critical dependence.

In practice, several directions could be reinforced.

1. Multi-vendor as the new standard of prudence

For many companies, the first response will not be an immediate shift to an entirely local solution, but a diversification strategy. Multi-vendor, already discussed in cloud, could become a more central principle for AI. The idea is simple: avoid having a single player control the entire value chain of a critical use case.

Concretely, this can mean:

• designing applications capable of switching models or providers;
• avoiding overly deep dependencies on proprietary APIs;
• maintaining several deployment options, including on distinct infrastructures;
• reserving certain sensitive use cases for more controlled environments.

This approach does not eliminate dependence on American players, but it reduces its intensity. It also turns sovereignty into an architecture issue, not just a political slogan.

2. The return of local solutions or hosting in controlled environments

The second possible consequence is renewed interest in models that can be deployed in environments more controlled by the user, whether dedicated hosting, local cloud, or infrastructure operated according to national or European requirements. The sovereignty debate generally pushes organizations to distinguish between generic uses, for which dependence is tolerable, and critical uses, for which it is much less so.

This distinction could become structuring in Europe. Public administrations, operators of vital importance, large industrial groups, or defense players have particular reasons to want to limit exposure to an external cutoff. Even if local solutions are sometimes less performant or more costly, the calculation can change once geopolitical risk is taken into account.

3. Increased pressure to build a European value chain

The concern expressed at the G7 can also strengthen the legitimacy of European industrial policies in AI. If dependence on foreign services is seen as a strategic risk, then support for a local value chain gains coherence. This concerns not only models, but also infrastructure, compute, software integration, and service offerings.

The issue is particularly sensitive for France, which has for several years sought to position itself in AI, both in research, startups, and infrastructure. In the French-speaking context, the question of sovereignty is not limited to language or regulatory compliance. It touches on the ability to offer solutions that can be used by public administrations, regulated groups, and SMEs without excessive dependence on outside platforms.

For French companies, this dynamic can create double pressure. On one side, they want access to the best global tools to remain competitive. On the other, they must meet growing requirements for data control, compliance, and resilience. The G7 debate reinforces the idea that they will have to arbitrate not only between cost and performance, but also between performance and autonomy.

Comparisons, the cloud precedent, and market repositioning

The strength of the issue also lies in the fact that it does not emerge in a vacuum. The AI debate takes up concerns already observed in other digital segments, especially cloud. For years, Europe has questioned its dependence on non-European providers for hosting, managed services, and critical software layers. AI does not replace that debate: it intensifies it.

The comparison with cloud is useful because it shows a recurring pattern. At first, the most technologically advanced players capture most demand thanks to their investment power, product maturity, and ecosystem. In a second phase, as their services become central, customers discover the strategic cost of that centrality. That is when demands for portability, reversibility, localization, and diversification emerge.

With AI, this cycle seems to be repeating faster. The market quickly adopted massively distributed American tools, often in the form of integrated services. But geopolitical awareness is arriving almost at the same time as adoption. That is what the warning relayed by TechCrunch reflects: leaders are not challenging the usefulness of these technologies, they are worried about the relationship of dependence they create.

The comparison with competing announcements is also illuminating, even if it must remain cautious. The entire sector now talks about “enterprise” AI, secure deployments, data control, governance, and compliance. But the G7 debate adds a dimension that product announcements alone are not enough to resolve: the political sovereignty of access. A provider may promise a high level of security, confidentiality, or compliance; that does not fully answer the question of who keeps the final say over service availability.

In other words, the competition no longer pits only models or platforms against one another. It also pits regimes of trust against one another. Companies and states will not choose solely according to the quality of generated responses, speed, or price. They will also have to assess the legal framework, geopolitical exposure, and the ability to retain room for maneuver in the event of a crisis.

For the European market, this could trigger a gradual repositioning:

• large American providers will remain very attractive on performance and ecosystem;
• European players will be able to differentiate themselves more on control, regulatory proximity, and resilience;
• integrators and consulting firms will have a growing role in designing hybrid architectures;
• public and private buyers could strengthen their requirements for reversibility and technical independence.

France is well placed to observe this shift, because it combines strong political ambition on digital sovereignty with a fabric of companies that already make broad use of American solutions. In practice, many organizations will not be able to do without models and infrastructure from the United States quickly. But they can seek to reduce lock-in points, better map their dependencies, and avoid placing all their critical uses in a single environment.

The debate over the “kill switch” could thus transform the way executive teams arbitrate their AI investments. Until now, the issue was often driven by innovation, productivity, or digital transformation. From now on, it could move more clearly to the level of strategic risk, with implications for procurement, security, legal, and data governance.

For French and European companies, AI sovereignty becomes a business continuity issue

Seen from the French-speaking market, the main interest of the warning relayed by TechCrunch is probably this: AI sovereignty stops being a theme reserved for states and becomes a very concrete issue of business continuity. A company that deeply integrates an AI service into its production, customer support, document management, or internal tools must now ask what happens if that service becomes inaccessible, limited, or politically sensitive.

This question is particularly pressing for regulated sectors and critical players. But it also concerns SMEs and mid-sized companies, which often adopt AI solutions through SaaS tools without always measuring the depth of the dependence created. The easier a use case is to activate, the more the risk of lock-in can be underestimated.

In this context, several questions should gain importance in tenders and technology audits:

• can data and flows be moved easily?
• can the application operate with several models?
• is there a fallback plan if a provider becomes unavailable?
• are critical uses separated from more generic uses?
• is the cost of changing provider known?

The consulting and integration market in France and Europe could benefit from this shift, because companies will need support to turn implicit dependence into explicit strategy. Sovereignty is not decreed only by choosing a national provider; it is built through architectures, contracts, governance policies, and use-case trade-offs.

For European public authorities, the G7 message could also serve as an argument for accelerating certain policies. The risk of an American “kill switch” gives a stronger political basis to investments in local capabilities, support for the European AI ecosystem, and control requirements for certain sensitive sectors. Here again, the key point is not to cut oneself off from American innovation, but to avoid its becoming a dependence with no way out.

It is nevertheless necessary to remain clear-eyed: building a credible alternative will take time. The United States retains a very visible lead in models, platforms, and infrastructure. European companies will therefore continue, in many cases, to use American solutions. The challenge is not rapid decoupling, but a methodical reduction of vulnerability.

That is precisely what makes the current moment important. As long as AI was seen as just another tool, dependence remained politically manageable. But if it becomes an essential layer of the digital economy, then the question of control over access can no longer be secondary. The debate raised at the G7 shows that this intellectual shift is underway.

The point raised by several leaders, as reported by TechCrunch, is not that an American provider is about to cut off access to global AI. It is that the mere possibility of that power becomes, in itself, a strategic problem.

In the long term, this awareness could reshape the market. The highest-performing providers will remain dominant as long as local alternatives do not offer a sufficient level of quality, integration, and cost. But the value of control will probably increase. In the coming years, offerings capable of combining performance, controlled hosting, interoperability, and reversibility guarantees could gain importance, especially in Europe.

For France and the continent, the window is narrow but real. If concern around an American “kill switch” is confirmed as a structuring theme, sovereign AI could stop being a niche segment or a political slogan and become a standard criterion in technology decision-making. The market will not shift overnight. However, each statement of this kind brings Europe closer to a scenario in which sovereignty will no longer be set against innovation, but treated as a condition of its durability.