When Progress Gets Switched Off: AI as a Question of Digital Sovereignty
The moment dependence becomes visible
Many will have noticed that the new model, Anthropic Fable 5, arrived with a doubling of the price. The price was high, and yes, that was fair to question. But the quality was exceptional. On some tasks the model didn't feel like a small upgrade, but like a genuine leap forward.
Shortly afterwards, access for users outside the US was apparently switched off.
And that is exactly where the real problem lies. This isn't about a single model. This scenario shows us Europe's dependence on technologies we don't control.
From a technology question to a sovereignty question
We have largely slept through the development of leading AI models. The major platforms don't come from Europe. The most capable hardware doesn't come from Europe. And even when it comes to applying AI in companies, we are often too slow.
Until now the convenient answer was: then we'll just use the best models from the US. But what happens when that access is restricted politically? Then AI is no longer merely a technology question. Then AI becomes a question of strategic sovereignty.
An old pattern reaches AI
The pattern is not new. We know export controls from the Cold War, from the chip industry and from the history of encryption. Powerful technology has always also been a geopolitical lever. Now this logic is apparently reaching the most capable AI models.
For Europe that is a warning signal, no reason to panic, but a reason to assess our own position soberly.
Europe's realistic opportunity lies in application
Our short-term opportunity probably does not lie in building our own frontier models on par with the US tomorrow. That would be desirable, but it is not a realistic immediate strategy.
Our opportunity lies in application. We can make available models consistently productive, including free and open models, many of which currently come from China and run on American hardware in European data centres.
That is not perfect, and honesty requires naming it:
- Open models can lag a few months behind the frontier.
- There are legitimate questions about transparency, training data and political influence.
- And no one guarantees that access to future hardware will always remain a given.
Sovereignty therefore does not mean naively relying on a single source, but deliberately managing dependencies and staying able to act, exactly as with cloud and supply chain.
The real bottleneck: we use the AI we have too timidly
The decisive point is this: most companies are not even beginning to exhaust the potential of today's AI. AI is often used like a better chatbot: a bit of text, a bit of summarising, a bit of research.
The real leverage only emerges when AI is integrated into processes:
- in compliance and evidence management,
- in IT security and threat analysis,
- in software development,
- in analysis, documentation and knowledge management,
- in decision support.
Europe is losing not only because we lack the strongest models. Europe is also losing because we use the models already available to us too slowly.
What mid-sized companies can do now
The good news: companies can change this, without waiting for the perfect, fully European and regulatorily settled solution.
- Make dependencies visible. Which AI services are already business-critical today? What happens if a provider changes price, availability or access?
- Plan for replaceability. Design architecture and processes so that switching models is not a full rebuild: an abstraction layer instead of vendor lock-in.
- Build in governance, don't bolt it on. Anyone using AI in compliance and IT security needs clear responsibilities, data flows and controls. Otherwise you create new risk instead of value.
- Start with a real process. Not “AI in general”, but one concrete, measurable use case with a clear owner.
This is exactly where we come in at Cybervize: AI not as a gimmick, but as a controllable building block in security and compliance processes, with the governance needed to ensure speed doesn't become loss of control.
Conclusion: sovereignty comes from acting, not from waiting
We don't have to wait until everything is perfect, European and regulatorily fully resolved. We have to start.
Because in the end, it isn't only the one who owns the strongest model who wins. It's the one who most consistently translates AI into real value and stays sovereign enough not to depend on a single point of access.