Fable is back online. The 19 days it was gone are the part to remember.

As of 1 July 2026, Claude Fable 5 is available again after US export controls were lifted. For most of June, one of the most capable models available was simply off, not because of an outage or a bug, but because of a government decision.

Anthropic returned under a condition: proactively detect and address security risks, and work with the government on standards for future models. The return is real. So is the reminder that the terms of access are set by parties who are not you and not your vendor.

Every continuity plan accounts for outages, rate limits, and price changes. Almost none account for a model becoming legally unavailable to your workforce overnight. For 19 days, that is the risk that bit.

The difference matters. An outage is measured in hours and has a status page. A policy revocation has no ETA, no SLA, and no support ticket that resolves it. It ends when a government decides it ends.

If a model is infrastructure for your business, treat it the way you treat any critical dependency you do not own: know the blast radius if it disappears, and have a path that does not require it.

For work that can tolerate the risk, hosted frontier models remain the right call. They are the most capable and the fastest to adopt. For work that cannot, the question is whether the capability runs somewhere you control. An open-weight model deployed on your own infrastructure does not get switched off by a letter. That is the point of sovereign deployment, and June made the argument better than any vendor deck could.

You do not need to move everything on-prem. You need to sort your AI-dependent systems into two buckets: those that can survive a sudden loss of a hosted model, and those that cannot. The second bucket is your sovereign-deployment roadmap.

The 19 days Fable was dark were a free preview of the day you find out which bucket a system was in.