VivaTech 2026 — Christel Heydemann Keynote

[00:00:00] Good afternoon, everyone. [00:00:07] Let me start with something very concrete. [00:00:11] A few days ago, access to Fable 5, the super powerful LLM from Entropic derived [00:00:19] from the highly publicized mythos, was suspended for non-US people [00:00:24] and companies in the name of US national security. [00:00:30] I think we should say it clearly. [00:00:32] This is not just another episode in the fast-moving AI story. [00:00:38] It is a wake-up call. [00:00:40] Because overnight, what many people still viewed as an open technological race [00:00:46] suddenly looked very different. [00:00:48] It reminded us that in AI, access is not always guaranteed. [00:00:54] It can be restricted. [00:00:56] It can be conditioned. [00:00:58] It can be decided elsewhere, especially outside Europe. [00:01:01] And when that happens, the question is no longer who has the best model, [00:01:08] whatever it means, the question becomes who remains in control. [00:01:14] For Europe, for governments, for companies, this changes the conversation. [00:01:20] It forces all of us to revisit some of our assumptions and, frankly, some of our priorities. [00:01:27] For years, the debate around AI has been dominated by performance, bigger models, faster adoption, more use cases, more productivity. [00:01:40] Of course, these things matter. [00:01:46] But what the Mythos episode reveals is that performance alone is not a strategy. [00:01:50] If access to a critical model can be suspended from outside Europe, then AI is not only a matter of innovation. [00:01:58] It's also a matter of geopolitics, strategic autonomy, resilience, and ultimately trust. [00:02:04] It means that both states and businesses need to ask harder questions. [00:02:10] Not only how fast can we deploy AI, but also on what foundations, under whose rules, with what guarantees. [00:02:21] Because a company that builds its transformation on tools it cannot fully access, explain, govern, secure, or economically control is not accelerating safely. [00:02:33] It's increasing its dependency and dependency in the edge of AI is a strategic risk. [00:02:41] And because the economic and power states, stakes are so high that promises alone have little values. [00:02:50] So today, I'd like to focus on one simple question. [00:02:54] How do we regain control without turning our backs to progress? [00:02:59] I do not believe the answer is to slow down or to step back from AI. [00:03:05] Quite the opposite. [00:03:07] The answer is to build AI on foundations we can trust. [00:03:10] First, we need to stop thinking about AI as something abstract, almost floating around us. [00:03:19] AI is physical. [00:03:21] It runs on data centers, network, cloud, environments, interconnections, energy systems, cyber security layers. [00:03:29] Talent and operating models. [00:03:32] An AI agent used by thousands of employees does not only rely on a model. [00:03:38] It relies on secure connectivity, cloud capacity, identity management, data governance and constant monitoring. [00:03:47] If one of these layers fails, the service may still appear available, but trust is already broken and the consequences can be dramatic. [00:03:58] And it also runs within political, legal and economic boundaries. [00:04:04] The mythos decision made that visible in the clearest possible way. [00:04:09] Second, regaining control means treating infrastructure as strategic. [00:04:16] At Orange, this is not only theory. [00:04:20] It's our daily reality. [00:04:22] We operate at scale across 26 countries and serve 340 million customers worldwide daily. [00:04:29] We rely on fixed and mobile networks, terrestrial and submarine infrastructure, satellite capacities, secure cloud environments and cybersecurity expertise to keep critical digital systems running safely every second of every day. [00:04:45] And this matters for AI because there's no trustworthy AI without trusted infrastructure. [00:04:54] Third, controlling the model itself is not enough. [00:04:58] You also need to control the access rights and that changes everything. [00:05:03] What matters is not only whether a model is powerful. [00:05:08] What matters is whether you can access it when you need it, whether you can deploy it under predictable conditions, whether you can rely on it over time, whether you have a deep understanding of its strengths and weaknesses and of its potential biases. [00:05:26] And strategic autonomy is also an economic issue. [00:05:30] AI is increasingly consumed as a service request by request, token by token for models whose economics are not yet fully stabilized for many businesses. [00:05:43] Many companies have experienced this very concretely. [00:05:48] A use case starts small. [00:05:51] Adoption grows fast and suddenly the bill grows faster than the value created. [00:05:58] Not to mention the societal and environmental responsibility with of such losses of control. [00:06:07] If you do not understand what drives cost of AI, if you cannot connect usage to value, if you cannot anticipate the economics of your dependency, then you're not fully in control of your transformation. [00:06:21] And finally, there's another dimension where this loss of control becomes even more serious. [00:06:30] Cyber security. [00:06:31] Because AI is not only changing productivity. [00:06:35] It's changing the nature of cyber risk. [00:06:38] The challenge is not only that attacks may become more automated and powerful. [00:06:44] The deeper challenge is that advanced models can identify chains of vulnerabilities, weak points and systemic flaws at a scale and speed that human teams alone will struggle to match. [00:07:01] That changes the rhythm of defense. [00:07:04] And when the pace of discovery accelerates, the ability to understand, prioritize, remediate and recover becomes even more critical. [00:07:15] This is why for us, cyber security is not a separate conversation from AI. [00:07:21] It's one of the places where the truth of AI will be tested. [00:07:26] If you cannot understand your own exposure, if you cannot investigate your own systems, if you cannot secure your critical environments without depending entirely on external black boxes, then you're not in control. [00:07:41] And without control, there is no trust, which brings me to my strongest conviction. [00:07:48] The more AI advances, the more strategic the human becomes. [00:07:54] We often hear that AI will replace people. [00:07:59] I believe the opposite is true. [00:08:02] The more machines produce, the more value shifts toward what they will never do. [00:08:09] Judgment, nuance, the courage to decide under uncertainty, the empathy, the intuition of a network engineer facing a complex failure. [00:08:23] The responsibility of a leader. [00:08:26] The company that can no longer write, analyze or decide without AI is not an augmented company. [00:08:34] It is a vulnerable one. [00:08:36] The next frontier of AI is not only technical, it is human. [00:08:42] Not because machines are not powerful enough, but because I am firmly convinced that values, judgment, courage and responsibility are not features that we can just download. [00:08:55] Staying in control does not mean rejecting AI. [00:08:59] It means refusing dependency and ease. [00:09:03] And that intellectual dependence, independence, or perhaps we should say that independence of judgment rests on physical independence. [00:09:14] Control of your data. [00:09:16] Control of your networks. [00:09:18] Control of your critical infrastructure and your compute capacity. [00:09:22] But above all, it rests on control of the models themselves and of the rights attached to them. [00:09:29] And for Europe, I believe this is now the real frontier. [00:09:34] Not simply building more AI. [00:09:36] But building AI that we can access, govern, secure and challenge on our own. [00:09:42] Moving together faster and with direction is now required. [00:09:46] While remaining open to the rest of the world, but without naivety. [00:09:51] At Orange, that's how we see our role. [00:09:54] We believe the next phase of AI will be defined by the ability to combine technological power, strategic autonomy, cybersecurity and human judgment. [00:10:06] We believe we have the opportunity to think long term and not always overreacting to external challenges. [00:10:13] That's how we will create lasting trust. [00:10:17] That's how we will create resilience for us, for our partners, for our clients and for our customers. [00:10:24] And that's how we will keep AI a tool for open progress, not a technological weapon to impose us a political and economic supremacy. [00:10:34] So yes, the mythos episode and it is a wake up call, but maybe it's also a very useful one that must make us stronger because it forces us to ask the right question at the right time. [00:10:50] Not only what can AI do, but what must we continue to control? [00:10:55] At Orange, our answer is clear. [00:10:58] We will use AI to think better, protect better and connect better. [00:11:04] But we will not let it sink in our place. [00:11:07] We will help build the infrastructure, the security and the human framework that trustworthy AI requires. [00:11:15] And that is why we trust the future. [00:11:18] Thank you.