The AI that vanished for 19 days is back. Here is the lesson.
On 12 June the most capable AI on the planet went dark, worldwide, on a government order. On 1 July it came back. If any part of your business leans on one AI tool, this one is worth two minutes.
Imagine your best supplier calling to say they cannot serve you any more, effective immediately, no end date. Not because they went bust. Because the government told them to stop.
That is roughly what happened to businesses around the world last month, except the supplier was an AI model. It had never happened before. It has now, and it is worth understanding, because it tells you something about how to build with these tools from here on.
What happened
Anthropic launched Claude Fable 5 on 9 June, and for three days it was the most capable AI you could get your hands on. Then the US government stepped in and placed export controls on it, the kind of restriction normally reserved for weapons parts and chip making equipment.
Because Anthropic could not instantly prove which of its users were outside the US, it did the only thing it could and switched the model off for everyone, everywhere. For 19 days the best AI in the world simply did not exist as a product.
On 30 June the Department of Commerce lifted the controls, and on 1 July Fable 5 came back, on the apps, the API and the big cloud platforms. As if nothing had happened. Except something had.
The trigger was a jailbreak, a prompt that talks a model into ignoring its own safety rules. Researchers at Amazon reportedly found one that could get Fable 5 to help identify software vulnerabilities, the raw material of cyber attacks. That report reached the government, and the government reached for a law built for physical exports. Applying it to a software model was a first, and it is why the blackout was global rather than tidy. You cannot geofence a model overnight when you cannot verify where every user sits.
The part worth noticing
The story is not really about one model or one jailbreak. It is that AI has quietly crossed a line, from clever software into infrastructure. Governments now treat a model launch the way they treat a shipment of sensitive hardware. That is new.
And infrastructure can be switched off. Businesses mid-project on 12 June found that out the hard way. Not a crash, not an outage with an apologetic status page and a fix by morning. A legal stop, with no promised end date.
What this means for a normal business
The wrong lesson is to avoid AI, or even to avoid that model. Fable 5 is back, it is excellent, and the whole episode arguably shows the safety system working, a problem was found, access was paused, the problem was dealt with.
The right lesson is about where you keep the value. If your prompts, your instructions, your data and your checks live in your own documents, then a tool disappearing is an afternoon of rewiring. If all of that lives inside one AI product, welded in, undocumented, then a tool disappearing is a crisis.
The businesses that shrugged off the blackout were not the ones that predicted it. They were the ones whose process was portable.
So what should you do this week
Write down the jobs AI actually does in your business. For each one, ask the awkward question, if this tool vanished tomorrow, what breaks?
Then do two cheap things. Copy your prompts and instructions into your own documents, somewhere you control. And run your most important AI job through a second model, once, just to prove you can. Twenty minutes, and the next blackout becomes someone else's drama.
Where we come in
This is how we build at Creative Sauce AI. The thinking, the prompts, the checks, the process, all designed to live with you, not inside any single tool. Models will come, go, and occasionally vanish by government order. Your business should barely notice.
If you are not sure what would break in your business if one AI tool vanished tomorrow, that is exactly the conversation worth having before it matters.
Book a quick chat →Related: The AI you can actually use just got a big upgrade
Common questions
Why was Claude Fable 5 switched off?
Three days after its launch on 9 June 2026, the US government placed export controls on the model, reportedly after researchers at Amazon found a prompt that could get past some of its safeguards and help identify software vulnerabilities. Because Anthropic could not immediately verify which users were outside the US, it suspended access for everyone, worldwide.
Is Claude Fable 5 available again now?
Yes. The US Department of Commerce lifted the export controls on 30 June 2026, and Fable 5 returned worldwide on 1 July, across the Claude apps, the API and the major cloud platforms. The shutdown lasted 19 days in total.
Has a government ever switched off an AI model before?
Not like this. Export controls are usually applied to physical things like weapons components and chip making equipment. Applying them to an AI model, and having the maker pull it offline globally as a result, was a first. It is the clearest sign yet that governments now treat frontier AI as critical infrastructure.
What should a small business do to protect itself?
Keep the value in your process, not the tool. Store your prompts, instructions and data in your own documents rather than inside one AI product, and test your most important AI job on a second model at least once. If a tool vanished tomorrow, switching should be an afternoon of work, not a crisis.