By: Emilie Pierre-Desmonde – Show Director
AI is no longer a subject for demonstration. It becomes an infrastructure layer. In companies, it is moving from “proof of concept” towards devices that really work, with users, risks, budgets. In the public sector, it is present in services where trust is not a bonus but a condition. And on the research side, the center of gravity is already shifting: we are talking less about “generating text” and more about systems capable of understanding, predicting, interacting, in short, of holding up in the real world.
In production, the criteria change. We no longer judge AI by a brilliant demo, but by very down-to-earth things: acceptable latency, a calculation bill that does not go off the rails, traceability, an ability to explain what happened, to correct quickly, to resist attacks and abuses. It is often at this moment that the debates become clearer: it is no longer just a question of models, but of operation, architecture, security, governance.
Why the WAICF particularly resonates in 2026.
It is in this phase that the World AI Cannes Festival (WAICF), on February 12 and 13, 2026 at the Palais des Festivals in Cannes, takes on a particular resonance. An event is never “the solution” alone. On the other hand, a major meeting can serve a very concrete purpose: to bring together actors who, during the rest of the year, often advance in parallel: researchers, industrialists, public decision-makers, field operators. And bring out in two days a more shared understanding of what works, what is blocking, and what deserves to be accelerated.
The closing keynote of Yann LeCun fits exactly into this moment. By pushing for an AI less focused on text generation alone and more oriented towards world model-type approaches (V-JEPA), it brings to the forefront a simple question: what is needed for AI to become truly reliable outside the laboratory? Behind the scientific discussion, there is a very operational signal: the next wave will also be played out on architectures, calculation, perception, and the robustness of systems.
Get your badge
Vision, constraints and production: a coherent program.
The rest of the program is built in this logic: bringing a vision into dialogue with evidence, and especially with constraints. This can be seen in the way the registers respond to each other. The joint ESA–Mistral AI keynote brings sovereignty back to practicable choices. Deployment, ability to operate under constraint, continuity, control of dependencies. It shows what it means to move from promise to exploitation. The technological trajectory can also be read in what we manage to hold in production: this is the whole point of speaking out like that of Robin Rivaton (Stonal) on the real impact of adoption. And because “real mode” AI mechanically expands the risk surface, the security dimension, supported by Wendi Whitmore (Palo Alto Networks), finds a natural echo in discussions on trust and governance, in particular with Leonardo Cervera Navas (EDPS) or Francesca Rossi (IBM). This dialogue between vision, implementation, security and governance is valuable: it sticks to the real trade-offs that organizations encounter.
A territory, uses, evidence.
Finally, it is impossible to ignore what the reality on the ground says. With the Alpes-Maritimes Department as a major partner, and the dynamic Alpes-Maritimes Terre d’IA which highlights its startups, Cannes also becomes a very concrete showcase of a French ecosystem which does not need to overplay: it advances, it builds, it delivers. And this is exactly what we expect from an AI that goes into “real mode”.
A lounge, at the back, is used to facilitate useful conversations, even the least photogenic. This is what the WAICF allows: to take the pulse of AI in 2026, but above all to leave with clearer benchmarks for the next stage, the one where AI is no longer commented on from afar and begins to be used as a system.
Get your badge
