Last week we got to spend a few days working with Anthropic's newest model: Claude Fable. And what an experience.
A bit of context. Fable was the publicly available variant of Mythos — the most powerful model Anthropic has built so far. Mythos turned out to be so good at cybersecurity and at finding vulnerabilities in software (some undiscovered for decades) that only a handful of selected companies and governments were given access. Fable is essentially the same engine, but with a solid extra safety layer around it — precisely so it *could* be used more widely.
And you felt that power immediately.
Last Friday I ran an experiment. A part of our software that a computer-science graduate had worked on for more than six months, I had Fable build. Within one hour there was a feature extension.
That's the kind of moment where you lean back in your chair for a second. Wow. This is genuinely next level.
And right at that moment a thought creeps in: *uh-oh.* Because the impact of this is enormous. When you deploy technology that is so much more powerful than what a programmer could deliver until recently, something fundamental shifts. On my own I can do things that were simply unthinkable a year ago. That's wonderful — and at the same time it raises the question of how we will organize our work going forward. In IT especially, that impact is huge.
But this morning a second *uh-oh* arrived.
I woke up to the news that the US government had, through an export-control directive, prohibited Anthropic from making Fable and Mythos available to non-Americans — inside and outside the US, right down to their own non-American employees. The result: Anthropic could not shut access selectively and had to switch the models off for *all* customers. Anthropic calls it a misunderstanding and says it is working on restoring access. The less powerful models, such as Opus 4.8, keep working as normal.
And there we are. Back to the models from before Fable — not bad, but a fair bit less powerful.
That gets me thinking. Because what is happening here is much bigger than one model going offline temporarily.
A government can, with a single stroke of the pen, shut off access to groundbreaking AI. Today it is fortunately "only" one model. But the direction is clear: the moment truly powerful models appear, the Americans want to protect their lead. And we in Europe are then left empty-handed.
How do we, as Europe, deal with that when we can no longer count on the technology our companies, our customers and our competitiveness come to depend on?
Is this the moment when we all need to look seriously at the European alternative, Mistral? Yes, the frontier currently lies in the US. But digital sovereignty is no longer a luxury — it has become a precondition. And a dependency that can be cut off from one day to the next is not a dependency you want as an entrepreneur.
I don't know the answer yet. But the question has definitely become serious.
Let me know what you think.
