The AI model you depend on can disappear in an afternoon

What happened

On June 12, 2026, at 5:21 pm Eastern, Anthropic received a US government export control directive. It barred any foreign national, anywhere, from accessing two of its newest models, Fable 5 and Mythos 5. To comply, Anthropic switched both models off for all customers. Its other models were not affected.

Anthropic's published account says the government gave no specifics, and that it understood the concern to be a narrow jailbreak technique. It said it disagreed with the decision and the process, apologized for the disruption, and said it was working to restore access. The US Department of Commerce lifted the controls on June 30, and access came back on July 1.

We aren't taking a side in the policy argument. The point is what it meant for anyone who had built on those models. One afternoon a dependency worked. The next, it didn't, for reasons no customer could influence.

So what for your business

Most teams think about model risk as accuracy and cost. This showed a third kind: availability. It can come from regulation, an outage, a pricing change, a contract dispute or a retirement notice. Providers retire models on a schedule, so even the quiet version of this happens to everyone eventually.

The useful question is not "will our vendor fail?" It's "how long are we down if they do, and what does that cost us?"

If the honest answer is "we'd need weeks to switch", that belongs in front of leadership, the same way a single supplier for a critical part would.

What to do this month

1. Put a seam around the model. One internal interface that every feature calls. Swapping a provider becomes a configuration change, not a rewrite.
2. Keep a second model warm. Pick a fallback, test it on your real tasks, and know what quality you lose.
3. Build an eval set. Fifty to a hundred real cases with expected outcomes. It's what lets you switch in a day instead of a month, because you can measure the result.
4. Read your data and retention terms. Some deployments carry retention requirements. Know yours before you need to change provider in a hurry.
5. Write the "model is down" runbook. Who decides, what degrades gracefully, and what the customer sees.