Anthropic ban and what it means for AI builders (in EU)

Models can come and go, we need redundancy

I just recorded a video about the Anthropic Fable/Mythos access story, and why it matters if you are building AI products in Europe.

Not because everyone should panic.

Not because we should stop using US models.

I still use them every day.

But the story makes one thing painfully clear: having access to a model is not the same thing as having control over your product.

AI in production needs reliability

If your product, internal workflow, or customer-facing AI feature depends on one frontier model provider, then you are also depending on their capacity, pricing, policies, routing decisions, retention rules, regional availability, and whatever export-control regime they operate under.

That used to feel like a theoretical risk. Now it feels much more real.

The question I am asking myself is rather simple:

What breaks if this model is not available tomorrow?

In a boring operator way I ask myself:

• Does the onboarding flow fail?

• Does the sales assistant stop working?

• Does the customer support workflow degrade?

• Do we suddenly send sensitive data through a path the customer did not agree to?

• Do we even know which parts of the product depend on which model?

I break down a practical way, how I approach this in production ai systems. Solutions include things like abstracting away model selection (using open router or https://vercel.com/ai-gateway) and in the code having an interface for this switching.

Open source models hosted on 0 data retention servers will def. get a boost from this I believe.

And learning how to build productions systems with AI-model and AI-provider redundancy will be more important than it was so far.

Exciting times, new ways for building product to learn.

Have a great day everyone

Cheers

Niko