Anthropic’s announcement about suspending access to Fable 5 and Mythos 5 hit differently for me.

Not because the models failed.

Not because the infrastructure went down.

Not because the technology suddenly stopped working.

Access changed because of a regulatory decision.

As a founder building AI products outside the United States, I read it less as another AI news story and more as a reminder of a risk many of us already live with every day.

Our products, customers, and businesses increasingly depend on models that we do not control.

That is a strange place to be.

We spend weeks optimizing prompts, building workflows, improving product experiences, and shaping entire features around frontier models. Then one decision, made somewhere far away from our users and businesses, can change what we are allowed to use.

Whether you agree with the decision or not is not the main point.

The point is that AI access is now business infrastructure.

And infrastructure risk matters.

If your product depends entirely on one provider, one model, one policy environment, or one country’s regulatory decisions, then your product has a hidden dependency. It might not show up in your dashboard. It might not appear in your error logs. But it is there.

Thousands of builders around the world can wake up one day and discover that a critical piece of their stack is no longer available to them.

This is why I am becoming more bullish on open-source AI.

Not because closed models are not incredible.

They are.

Frontier models are still ahead in many areas. They are easier to use, more polished, and often much better for complex reasoning and product workflows.

But resilience matters too.

As AI becomes core infrastructure, access, ownership, and independence become just as important as intelligence. Founders need to think about fallback models, open-source alternatives, data portability, model routing, and what happens when a critical provider is suddenly unavailable.

The best model is not always the one with the highest benchmark score.

Sometimes it is the one you can still rely on tomorrow.