Sorbonne University expertise at the heart of European AI governance

Sorbonne University has just been recognised as a European expert in artificial intelligence. Can you tell us more about this appointment and what is at stake?

Arnaud Latil: To fully appreciate the stakes, we need to bear in mind how governance is structured under the EU Artificial Intelligence Act. This legislation stipulates that the AI Office, established within the European Commission, is responsible for part of the regulation’s implementation. To carry out its task and to inform its decisions, two advisory bodies have been set up. The first is a scientific committee of independent experts. The second is an Advisory Forum, and it is within this body that Sorbonne University has been selected to sit.

We work alongside other research organisations, as well as representatives from businesses, SMEs and civil society organisations. Our role is to advise and inform the AI Office on the practical implementation of the regulation. This is a ‘technical’ advisory role covering engineering aspects as well as organisational, ethical and legal considerations.
 

Sorbonne University is one of the two French universities selected to sit on this advisory forum. In your opinion, what made the difference? 

A. L.: We were selected alongside the University of Rennes (the Sequoia cluster), INRIA, the CNRS and the CEA. I believe the strength of our application lies in what Sorbonne University, which includes SCAI, has to offer: true diversity and a wide variety of expertise in AI fields.

Our expertise is not solely technical or mathematical; it extends to computing, law, organisation, sociology and even philosophy. We possess a large range of interdisciplinary skills, which is the very essence of our PostGenAI cluster, which has been recognised at a European level. The aim is to imagine what AI might look like beyond generative AI, for example in the field of agentic AI. 
 

What will your day-to-day role be, and the university’s role, within this advisory forum?

A. L.: Thematic working sub-groups will be set up. Although the final list has not yet been finalised, the main themes will focus on regulation, the creation and coordination of different norms (particularly for so-called ‘high-risk’ use cases), the implementation of regulatory sandboxes¹, and finally the coordination of AI governance between European and national levels. We will be focusing on these key topics, although it is still too early to say precisely which ones we will prioritise.

How do you intend to give French academic research a voice within this multi-stakeholder forum?

A. L.: This is a fundamental question. Obviously, our positions will depend on the specific issues put before us. However, the guiding principle is clear: the real challenge in implementing this regulation is to propose sustainable norms – with provisions on high-risk use cases due to come into force, in principle, from 2 December 2027. The aim is to ensure that the text is implemented in a way that is realistic, tailored to practical applications and usage, rather than resulting in regulations that are out of touch with real life.

The implementation of the AI Act’s requirements must be clear, relevant, appropriate and adaptable for businesses and organisations. It must be flexible enough not to hinder technological progress, yet structured enough to protect citizens. This is precisely why the advisory forum brings together research centres, think tanks and private companies. The regulator needs a range of expertise to consider multiple fields simultaneously. We aim to ensure that the theoretical requirements of the regulation are grounded in the reality of the economic and scientific landscape.

The AI Act is often criticised by some industry players who fear it will hinder technological innovation in the face of competition from American or Asian giants. How would you respond to them?

A. L.: This legislation was specifically designed to strike a balance between two objectives: encouraging innovation on the one hand, and putting safeguards in place on the other. Europe is taking concrete steps to support the development of AI, notably through provisions on computing power accessibility, regulatory sandboxes and, lately, through the pursuit of technological sovereignty in the areas of cloud computing and AI. But it is also setting limits and guidelines to ensure that the use of AI is consistent with the values of the European Union.

These requirements – which relate to human oversight, transparency, data governance, robustness and cybersecurity – strike me as essential for practising ethical AI. The real challenge lies in putting them into practice. Take the example of human oversight: who could possibly be against it? Yet what does it actually mean in a line of code or an industrial process? If the requirements are too rigid, they will stifle innovation. Conversely, if human oversight is insubstantial or a mere formality, we will fail to live up to the European values we claim to uphold. Finding the right balance is our mission.

How do you intend to mobilise the key players at Sorbonne University to contribute to the European AI Board?

A. L.: We could set up internal working groups to share relevant information – scientific, ethical and legal – depending on the topics being discussed in Brussels. For example, we can safely assume that a European working group will look at issues specific to AI in the healthcare sector. Our colleagues at the Faculty of Health would obviously be called upon to enlighten me so that I can provide cutting-edge expertise at a European Union level. Let us not forget that the Advisory Forum is a multi-stakeholder body comprising 174 organisations: to have a voice and be heard there, our contributions will need to be firmly underpinned by our research laboratories.

How will this appointment shape future research initiatives or new study programmes within the university?

A. L.: In the immediate term, this will have no direct impact on academic programmes or research budgets, as the forum’s primary objective remains to advise the European regulator. We are here to contribute our practical experience to research, teaching and even administration. 

However, this will provide our community with a valuable opportunity: a direct presence at the heart of European institutions and first-hand insights into the legislative processes underway. Our goal is to be part of global public debate. The project establishes Sorbonne University as a key player in public policy on AI. It enables us to advise European administration and represents an active contribution from the research community towards civil society through our expertise.

Advising decision-makers is not an isolated initiative in terms of Sorbonne University’s outreach to society, is it?

A. L.: Indeed, we are already pursuing this approach through other programmes, such as the SOUND innovation programme, which is strongly focused on societal challenges and science outreach.

Within SCAI, we are also conducting very practical joint research projects, such as a partnership with the Court of Cassation. Together with our colleagues at Université Paris-Panthéon-Assas, we are carrying out an experiment on the use of AI by the court’s advisers.

Our commitment to fostering dialogue between research, public administration and businesses is also reflected in our participation in major events such as VivaTech. Sorbonne University will be actively involved, and we will be organising a round table in partnership with the European Commission dedicated to a crucial topic: the transition from research to market in the field of AI. Our mission is fully aligned with this broader drive to disseminate knowledge, from our laboratories to public and private organisations, and to the wider economic and social worlds
 

^{1   Schemes enabling businesses, start-ups or researchers to test their technological innovations in real-world conditions, with relaxed regulations for a specified period.}