More centres of power.

Anthropic being forced to shut down Fable is not primarily an AI problem. Nor is it a US problem. It’s an example of what happens when the people calling for “more control over AI companies” get exactly what they wished for.

I struggle to see how moving that power to European or Swedish alternatives would solve it. Why would future French presidents show more restraint than Trump? And if Sweden, against all odds, built world-class models of its own (at a cost of a little more than the entire national budget), would we really want to put that power in the hands of whichever government happened to be in office?

To me, this points in a different direction.

For a long time, we’ve built democratic societies around the idea of separating powers. Church was separated from state. The media developed into an independent force of scrutiny. No single institution should be able to control everything.

Perhaps we should think the same way about AI.

1. We need a clearer separation between the guys with the guns and the guys with the GPUs.

Not because technology companies are always right. Not because states are always wrong. But because concentrated power almost always becomes a problem, whoever holds it.

2. At the same time, we need to keep strengthening open source, open weights and all kinds of open alternatives.

They will probably never be as powerful as the commercial models at the very front of the curve. But they are already good enough for a great many uses, and they provide an important counterweight to both state and commercial monopolies.

If this story shows anything, it’s why we need more centres of power, not fewer.