On 9 June, Anthropic released Claude Fable 5 — the most capable AI model it had ever made available. Three days later, the US government applied export controls and the model vanished. Globally. Mid-workflow, for everyone using it. It came back on 1 July, nineteen days later.
If you missed the whole saga, good — you weren’t depending on it. But the episode is worth five minutes of any business owner’s attention, because it’s the clearest demonstration yet of a risk most of us haven’t written down anywhere.
What actually happened
The short version: researchers found a way to bypass some of Fable 5’s safety guardrails, a report reached the US government, and under a new AI security order the model was pulled from the market while it was reviewed. The controls were lifted on 30 June and the model was redeployed the next day. Anthropic handled it about as transparently as you could ask.
No villain, no scandal. Just a new kind of weather: the most advanced AI tools now sit close enough to national security concerns that a government can switch them off, and this year one did.
The lesson isn’t “avoid AI”
Nineteen days is a long time if your quoting process, your inbox triage, or your reporting runs through one model. And this won’t be the last interruption — models get deprecated, prices change, terms change, and occasionally, apparently, governments step in.
The businesses that shrugged this off had something in common: their processes belonged to them, not to the tool. The AI was doing the typing, but the how — the steps, the checks, what “good” looks like — was written down in plain English and could be pointed at a different model in an afternoon.
Treat any single AI model like a key supplier: brilliant to have, unwise to depend on without a plan B you’ve actually thought about.
That’s the whole lesson. Not paranoia — portability.
A small, honest next step
Make a short list of the tasks in your business where AI is now load-bearing. For each one, ask: if this exact tool disappeared on Monday, what would we do? If the answer is “open a different AI tool and paste in the same instructions”, you’re fine — that’s a plan B. If the answer is “I don’t know, it’s all in Steve’s chat history”, spend an hour writing the process down. The nineteen days someone else lost can be the hour you invest instead.