Digital Sovereignty: Europe Faces the Challenge of Trustworthy AI

Digital Sovereignty: A Strategic Challenge for AI

Digital sovereignty has become a fundamental criterion for the deployment of artificial intelligence in regulated sectors. For a long time, this concept was viewed as a theoretical debate, holding little weight against other considerations such as tool performance or cost. However, with the advent of the Internet, cloud computing, and more recently AI, a dependency has developed on major American platforms like Microsoft, Amazon, Google, and now OpenAI or Anthropic.

This pragmatic approach, sometimes deemed naive, is changing. Geopolitical stakes, the concentration of critical technologies, and the growing importance of data—essential for AI and crucial for competitive advantage—have altered the perception of risks. Sovereignty is no longer just a principle; it is becoming an operational criterion. This transition, albeit partial, is made tangible by the European AI Act, which lays the groundwork for a more regulated use of artificial intelligence.

From Theory to Practice: Sovereignty in Action

In discussions, the desire to reduce dependence on foreign actors is widely shared. Companies and institutions are increasingly favoring European solutions, such as those offered by Mistral AI. However, in practice, initiatives like the "anti-Huawei" law on 5G have not always succeeded. The most effective, accessible, and integrated tools remain predominantly American, often relegating sovereignty to a secondary concern, decided on a case-by-case basis.

Sovereignty becomes truly non-negotiable in the most sensitive environments. In sectors like defense, healthcare, and critical data, the question of control takes precedence over performance. The project to migrate the Health Data Hub to SecNumCloud-certified solutions illustrates this trend.

AI Confronted with a Wall of Distrust

The tension between tool performance and sovereignty requirements is particularly evident in the field of artificial intelligence. Although AI usage is spreading rapidly, large-scale deployment remains limited in regulated sectors. According to the study "The State of Enterprise AI and Modern Data Architecture," 74% of companies identify security and compliance as major barriers to adopting AI in production. Meanwhile, the global cost of cybercrime could reach $10.5 trillion per year, equivalent to the third-largest economy in the world.

The main obstacle to the regulated deployment of AI in businesses is a persistent distrust. This primarily concerns the reliability of models, which are improving but still perceived as difficult to audit. Additionally, security raises significant concerns regarding data circulation. Finally, sovereignty is a major issue in a context where the dominant infrastructures and models are largely foreign.

These limitations have concrete consequences. Some organizations restrict access to AI tools, while others see the emergence of unregulated uses. The paradox is that AI is already being used, but it is not yet fully mastered. In sectors like healthcare, where AI is already well-established, particularly in image recognition or diagnostic assistance, the uses related to generative or agentic AI are just beginning to emerge.

Regulated Sectors: New Laboratories of Digital Trust

The healthcare sector is not an isolated case. Other heavily regulated fields, such as legal professions (notaries, lawyers, accountants), face the same limitations: AI cannot be deployed according to the same logic as in less constrained environments.

These activities serve as a testing ground for AI, where digital trust emerges as a prerequisite. It is not just about performance or productivity gains, but about legal reliability, decision traceability, accountability of actors, and compliance with regulatory frameworks.

It is precisely in these constrained environments that a new generation of AI is taking shape. This context opens up an opportunity. As trust becomes a condition for use, a new technological layer is emerging, focused on securing, governing, and mastering AI systems. Cybersecurity for agents, identity management, data flow control, and more sovereign hosting: these components, still fragmented, address a very concrete need to frame powerful technologies within constrained environments.

A Strategic Window for Europe, Still Fragile

Investments are beginning to be structured around these issues. In France, the "Trust Tech" sector is progressing despite a generally declining venture capital market. The average size of funding rounds has increased from around €7 million to €15 million in two years, with players like GitGuardian, Stoïk, or Tomorro. These dynamics are still scattered, but they reflect a growing awareness: value will not solely lie in the models themselves.

After an initial phase focused on models, then on infrastructures, a new layer is gradually asserting itself. This layer enables artificial intelligence to be usable in operational contexts, ensuring its reliability, traceability, and security. In other words, making it deployable.

This shift is part of a broader movement. Artificial intelligence is finally being recognized as a strategic issue. Just as with nuclear technology in its time, the United States has gained a decisive lead, but Europe is attempting to structure a response. The signals are there: players like Mistral AI or AMI Labs have successfully raised several hundred million euros, a level that was still unthinkable a few years ago in France.

However, a structural limit remains. While Europe is managing to bring forth new players, it still struggles to compete with American funding power and to support these companies over the long term. The question is no longer just about launching initiatives but about staying in the race. This is where the next phase of the market will unfold. Not by marginally improving model performance, but by making their use possible in demanding environments. And this also implies, in the coming years, making digital trust one of the main areas of technological investment.