I was using Anthropic's Fable 5 one day and the next day it was gone.
That experience made me think about a much bigger issue. AI strategy cannot depend on a single model or a single vendor. Access can change for many reasons: commercial decisions, regulation, export controls or geopolitics. If an important business process relies on one provider, that dependency quickly becomes a business risk.
This is where AI sovereignty starts to matter. Not just where models are hosted but whether organisations have the flexibility to adapt. Keep the model layer separate from business logic, use open interfaces where possible, and design applications so they can move between models without major rework.
The real advantage will not always come from using the latest model. It will come from building systems that can keep moving when the AI landscape changes.