So. The US has blocked export of
@AnthropicAI's Fable 5 and Mythos 5, and the noise around it is mostly grift. I keep coming back to this because the reality is orders more boring than the outrage, and I probably won't get the views or impressions for saying it. But truth is truth.
I tested Fable earlier this week. Got it to render a poetic translation of the opening stanzas of Pushkin's Eugene Onegin, and it did an absolutely cracking job. The slightly unnerving bit is that it's super capable across domains - think of it as a research scientist who also happens to be a multi-instrumentalist. The thing can probably do biological research capable of creating biological weapons, or developing other AI models. I'm not saying that's what it's for. And Anthropic tuned it conservatively, so it rejects tasks that smell like foul play. But that raw capability is probably there.
Here's the but. For Britain to prosper economically using AI, we simply don't need models like Fable.
We don't.
We're stuck in such a deep productivity rut, mostly in SMEs outside London, that flipping out because Fable is suddenly unavailable is like someone who can't ride a bicycle creating vitriol because they were denied a license to pilot a space shuttle to Alpha Centauri.
There are metric tons of other models - proprietary ones from Anthropic themselves,
@OpenAI,
@GoogleDeepMind in London, the French lab
@MistralAI, open weight and open source options - all capable of running 99.9% of business automation tasks, at a fraction of the cost of Fable or even Anthropic's Claude Opus 4.8, which wasn't affected anyway.
Look, if we step back from the sensationalism, the actual problem is elsewhere.
We have two of them, really.
First: we don't have enough capacity to run AI inference at scale inside Britain. That means moving electrons for every token that comes out of a model, and moving electrons requires energy that needs to be cheap. Britain's industrial electricity prices are almost 50% higher than France and Germany, about four times higher than the US and Canada. We're lagging behind even the EU.
Data centres need to run 24/7, and for a country that doesn't produce a lot of fossil fuels, the most surefire way to enable AI inference growth is nuclear - cheap, abundant, ideally co-located with the data centres. We're hopelessly behind on that.
Building reactors here requires bespoke design due to an inordinate amount of regulations. Standard reactors, not even small modular ones, are absolutely ridiculously expensive because of it. Just fixing that would uncork the supply of AI data centres and inference capacity, and yes, that would give us actual AI sovereignty - running models here, serving customers here - in case the rest of the world decides to shut off access, which they haven't done.
The second problem is where research actually gets applied. British universities crank out decent fundamental AI research. The IP just doesn't stick around. Partly policy - our spin-out equity structures discourage researchers and investors compared to the US. Partly because there's nowhere to run it at scale. You need abundant data centres and cheap, consistent backbone energy. We have neither.
I'll be straight. There's a kernel of truth in what Alex Armstrong is writing. AI sovereignty matters - we need our own steelmaking, our own food security, our own ability to make weapons without relying on external supply chains. We also need to build and run our own data centres for AI, no question. But in the short term, the opportunity for prosperity in Britain is applying existing models that have been developed elsewhere. And for that, we absolutely do not need super capable models like Fable 5.