AI sovereignty: the most expensive outage is the one you inflict on yourself
Late June, a MEDEF garden party. Between two glasses, a speaker you don’t quite expect in that setting: Asma Mhalla, there to talk about the geopolitics of AI and digital sovereignty to a room full of business leaders. My first reaction was irritation. Why invite someone who worries executives, when what the country needs is to accelerate its adoption of AI, full stop? I almost stopped at that impression.
Then I listened again. A second time, properly, opening my mind to everything she covers. And honestly, it took me a week to digest. Here is where I landed.
What first annoyed me: “everything is political”
Her through line is that technical choices are political choices, that AI is first and foremost a matter of power and geopolitics. It’s a brilliant hook. But it proves a little too much. Her reasoning: since the United States can cut off our access to their models, we must seek independence. Except that, taken to its logical end, the same argument says everything and its opposite. If the danger is that Washington cuts off our supply or attacks us, then no factory should have a single production site either. A formula that applies to everything ends up deciding nothing.
I have to grant her the finesse: she frames it strategically, never lapsing into brutal doom-mongering. And let me say it plainly, a mind of rare speed, a pace that makes you run to keep up. A talk that forces you to listen again is already a talk that counts. She has, in fact, been making this case for a good ten years: that the digital question is, above all, a political one. This is no read of the moment, it’s a long-held conviction.
But the dependency is very real
Where I followed her without reservation is on dependency. And the news was right on cue: a few weeks before her talk, in the wake of a US executive order making AI a matter of national security, access to Mythos, a particularly advanced cyberdefense model, was restricted to US nationals only. There, we leave theory behind. A supplier can shut the door overnight.
An industrialist gets that instantly, because it’s a risk he already knows by heart: the single supplier, the single source that holds you. Twenty-five years in plants taught me that in my gut, not on a slide. I have seen a line sit idle for several days after a cyberattack. The risk exists, it would be dishonest to deny it.
The real risk I see on the ground: paralysis
And yet. On the ground, what destroys the most value is not the attack. It’s fear. I see IT departments and cyber teams so restrictive that, true, there is no incident. But nothing happens anymore either. Innovation slows, projects die before they’re born, employees disengage. Value creation, meanwhile, is well and truly at a standstill.
The gap between the fear we now inject into everything and the real risk strikes me as enormous. And no one measures it, because an outage you inflict on yourself makes no noise. It is, however, the most expensive one.
The car doctrine
I repeat this image often. The surest way never to have an accident is not to drive. If we scared drivers the way we scare executives about AI today, no one would take the wheel. Yet we have, collectively, accepted the risk of the road, the accidents, the injuries, the cost to the community, for a higher value: the freedom to move about.
AI is the same trade-off. The role of good counsel is not to reassure by forbidding, nor to charge ahead by denying the dangers. It is to lay every option honestly on the leader’s table, including that of accepting slightly greater exposure to test frontier models, on the sole condition that he is fully aware of the stakes. That is what taking the wheel with your eyes open looks like: a clear-eyed strategy, not a fear neatly filed away in a binder.
Accelerate, but safely
My conclusion holds in two words that belong together: accelerate, safely. What is missing today is not reasons to stay still, we will always find those. It’s a standard that says who, in IT, has the competence to accelerate safely, rather than freezing everything out of caution. That is exactly what I am building with ANGLE 360.
The concrete move, for an industrial leader, on Monday morning: nurture the relationship with your IT director, because everything rests on them, they are the guardian of your IT security. But pair them with an external, independent perspective, to check that they protect without ossifying. The right question to put on the table: what are we not doing today, purely out of fear?
Thank you to Asma Mhalla for the jolt: a talk that forces you to listen again and then digest for a week is a rare thing. I don’t share all of it, far from it. But I will use it to strengthen Angle Intelligence, because the sovereignty that truly matters to a leader isn’t cutting yourself off from the world, it’s keeping your hand on your own choices.
For the full picture, read the guide AI in industry. See also: Which AI tool for which industrial use. Wondering where to start? Gauge your AI maturity in 2 minutes, or let’s talk for 20 minutes.