The Claude Fable 5 shutdown, AI sovereignty and national security

Published on

After the US government shut down Claude Fable 5 for any non-American citizen overnight, I couldn’t stop thinking about one thing: what does this mean for any country that runs critical infrastructure on third-party models via API?

The Claude Fable 5 shutdown, AI sovereignty and national security

The summary of the facts: Anthropic received a government order requiring it to block access to Fable 5/Mythos 5 for any foreigner, inside or outside the US, including Anthropic’s own employees, like Andrej Karpathy. Since doing that is technically unfeasible, Anthropic took the models offline for everybody.

This is a very clear warning that any country that depends on the models from the big American labs is subject to losing access to the best of the technology overnight, without any notice or explanation.

Right now, control over who uses the best AIs in the world is concentrated in a single point, in Washington, DC. Not in São Paulo. Not in Berlin. Not in Tokyo.

And that brings back the topic of AI sovereignty, which I’ve been talking about for a long time on IA Sob Controle.

And sovereignty has nothing to do with state-owned companies. France’s Mistral is an example of how this can work. France already swapped Palantir for domestic software, invested €655 million in national AI development and put Mistral models in the hands of public servants and in strategic areas of the country. In other words, it didn’t stay in the speeches: it put critical national infrastructure running on models that can’t be shut down by the US. In Brazil, we can mention Embraer in the aviation sector.

To have real AI sovereignty, any country or bloc needs:

  • Its own models, or at least access to open source models that run locally. Here, China is ahead with DeepSeek and similar ones, and not by chance: being under a US embargo, developing a local alternative stops being an option and becomes an obligation.

  • Its own GPU infrastructure and computing power on national territory.

  • National talent and data to train and adapt models.

One point I think deserves more attention in this discussion is SLMs. Unlike frontier models, which require giant datacenters to be trained and to run, SLMs are smaller and can be hosted locally by companies, at a viable computational cost, and even fine-tuned with their own data when it makes sense. They don’t replace the big models at everything, but for a good chunk of use cases they can do the job.

Brazil has been talking about this, mainly with the PBIA, but still very timidly. Our GPU infrastructure still isn’t competitive at a global level. I REALLY hope we get there soon.

I’m not saying we should drop Claude or GPT tomorrow. Far from it. But the Fable 5 case showed clearly that relying ONLY on third-party models is a risky bet for any country.

It’s a national security issue. The future will belong to whoever controls this technology, and we’re still far from it.