In a context of American and Chinese dominance in AI, data control is becoming a strategic imperative. Digital sovereignty, algorithm transparency and data security are emerging as the pillars of European technological autonomy, which has yet to be built. ‘Dare to use AI’ proclaims the government, but what lies behind the slogans?

In a context of American and Chinese dominance in AI, data control is becoming a strategic imperative. Digital sovereignty, algorithm transparency and data security are emerging as the pillars of European technological autonomy, which has yet to be built. "Dare to use AI" proclaims the government, but what lies behind the slogans?

By S&D Magazine

 

Risks to sovereignty posed by the emergence of AI

Paris, February 2025. The Summit for Action on Artificial Intelligence is causing quite a stir in the French capital, bringing together thousands of industry players from around 100 countries. Heads of state, international organisations and businesses gathered to discuss the future of AI and its regulation. Among the topics discussed were data sovereignty and control. When it comes to AI, American hegemony is indisputable.

In 2024, the US developed 40 AI models considered noteworthy, compared to 15 for China and... only 3 for the European Union1. AI exacerbates issues related to data sovereignty as it becomes increasingly prevalent in the strategic processes of businesses and governments. The reason for this is that AI training requires access to massive amounts of data, some of which is sensitive or even personal. This structural dependency largely benefits the American and Chinese giants that dominate the global ecosystem, creating a strategic vulnerability for European players.

To make matters worse, the legal framework reinforces this asymmetry, as the Cloud Act authorises US authorities to access data stored by US companies, even when it is hosted on servers located in Europe. This reality undermines the EU's ambitions for digital sovereignty. The lack of algorithmic transparency adds to this fragility, raising questions of transparency, trust and accountability.

Paris, July 2025. The government announces the "Osez l'IA" (Dare to use AI) plan, a national programme aimed at accelerating the adoption of AI in all national companies: large groups, mid-sized companies and SMEs. The ambition is clear: to make AI an accessible, practical and useful tool for all French companies, wherever they are located and whatever sector they operate in, by 2030. The programme includes an AI Academy, free training courses, AI data diagnostics, solution catalogues, state-guaranteed loans, AI ambassadors and business meetings. "The public authorities must now protect companies operating on European soil by requiring them to use a minimum percentage of European tools. The Americans do it, the Chinese do it, so why shouldn't we?" asks Homeric de Sarthe, CEO of Craft AI2.

Europe must take a stand

Europe no longer has a choice. "The fight for digital sovereignty is above all a political battle. The Paris summit on AI must be an opportunity for France to assert a clear vision. What do we want? To be a player or a spectator? Let us raise ourselves to the level of other world powers with an ambitious plan. The AI Foundation is capable of fulfilling this hope and potential, provided it focuses on these strategic areas and ensures full political support for their adoption. It is not too late to act, but every month of inaction creates a gap that is difficult to close. We have the necessary infrastructure and talent, as well as promising initial projects (Prometheus-X, EONA-X, TEMS, Pleias, LightOn, Mistral, Albert, Outscale, Giskard, CleverCloud, OpenLLM, etc.). What we need is strong political commitment," noted an opinion piece signed by several industry players, including Philippe Latombe, Member of Parliament for Vendée, and Bernard Benhamou, Secretary General of the Institute for Digital Sovereignty.

"If Europe fails to organise itself to control its dependencies, if the legislative and investment measures it has taken do not produce the expected results in the short term, its economy will face three systemic risks: geostrategic risk, economic risk and legislative risk. Sovereignty is therefore not a luxury, it is a strategic necessity. Our proposals are concrete actions and we are determined to contribute to their implementation"3 continues the French opinion piece.

Keen to take a ‘third way’ approach to the American and Chinese giants, Europe is seeking to position itself. As is often the case, it intends to do so through regulation of the sector, but also by promoting innovation and research to encourage the emergence of frugal, explainable AI (capable of justifying its decisions and preserving human autonomy) while maintaining data sovereignty. A few months ago, the EU announced €20 billion in investments to build five AI giga-factories that would enable the training of AI models far more complex than what current capabilities offer. This is a first milestone on the long road to digital sovereignty.

"European companies can specialise in creating and deploying specialised, high value-added AI, rather than trying to compete with an idyllic, global, generic and universal tool. But the infrastructure remains too small and entirely dependent on American equipment manufactured in China and incorporating Taiwanese chips" points out Homeric de Sarthe4.

Balancing AI and data control

Several strategies already enable companies to reconcile data control with the use of AI, as the uncontrolled use of these tools raises real issues.

Nearly one in two employees has already used AI in the course of their work without informing their superiors, and only 9% of French employees have access to an ethical charter concerning the use of AI and/or a point of contact for questions on the subject5. In fact, many organisations are adopting a hybrid approach, combining on-premise infrastructure to host the most sensitive data, so-called sovereign clouds for applications requiring greater flexibility, and public cloud services for less critical uses. This strategy makes it possible to adjust the level of security according to the sensitivity of the data, while optimising costs and operational agility. Retrieval-Augmented Generation (RAG) enables the generation of accurate responses by limiting the need to transfer data to external services. Data confidentiality is thus preserved and employees can use the AI platform in complete security.

Another technique is local fine-tuning. Particularly suited to specialised professions with precise technical terminology, local fine-tuning allows a pre-trained model to be adjusted using the company's data and the AI to be customised according to the specific needs of the business, while keeping sensitive data within the infrastructure and improving the model using the use cases provided.

A French hospital has developed an AI solution for analysing medical images. Using a locally fine-tuned open-source AI model with anonymised data, the Health Data Hosting certified infrastructure also relies on a RAG system for diagnostic support. The result: a 30% increase in diagnostic accuracy, shorter treatment times, and greater patient confidence in the use of their data.

In addition to health, education is also investing in this field. The ‘Data Mastery in the Age of AI for Education’ (AI-DL) project, launched in January 2025, aims to strengthen data mastery among teachers and students in the age of artificial intelligence through practical experience with AI solutions. Its main objective is to equip educators with the skills they need to navigate AI technologies effectively and ethically. The project is developing a data literacy framework and an online training course, integrating collaborative communities of practice in schools. These communities bring together teachers, school leaders and students to test innovative approaches. Their feedback will be used to guide education policy. Expected results for 2028 for the seven European countries involved: Spain, France, Ireland, Italy, Lithuania, Luxembourg and Slovenia.

The race for artificial intelligence is not just a matter of technological innovation, but is also a strategic issue of power, politics and economics.

"AI is seen as a black box with magical powers. Training people on AI and the possibilities it offers will help reduce fruitless debates. We need to focus on use cases and the benefits that can be gained from them. SMEs are the champions of agility. If they see it as a way to increase productivity, and therefore competitiveness, they will adopt the solution" says Homeric de Sarthe6. Let us hope that it will be European...

 

 

 

 

______

1Can EU’s €200B ‘InvestAI’ Initiative Close Europe’s AI Gap with U.S. and China?
2Osez l’IA : ce que pensent vraiment les pros du plan du gouvernement
3Tribune "La France doit aussi lancer un grand plan de l’IA pour que notre souveraineté numérique soit assurée"
4Osez l’IA : ce que pensent vraiment les pros du plan du gouvernement
510 chiffres clés sur l’utilisation de l’IA au travail
6Osez l’IA : ce que pensent vraiment les pros du plan du gouvernement