Vanished.

Three days ago, I wrote that Claude Fable 5 was the most powerful AI ever made generally available, and that the striking thing was the price: free for a fortnight, pennies a use after that. Go and try it, I said, while it costs nothing.

On 12 June, the US government ordered it switched off. Anthropic complied within hours. Not just Fable 5, but its restricted sibling Mythos 5 too, disabled for every customer in the world. The most advanced model on the market, gone three days after launch. Not beaten by a competitor. Not a bug. A letter.

What actually happened

The order came from the US Commerce Department, citing national security and export-control powers. It barred access to the models by "any foreign national," anywhere. Because Anthropic could not reliably tell, in real time, which of its users were foreign nationals, it had to switch the models off for everyone, everywhere. Its other models, including Opus 4.8, stayed online. (Anthropic set all this out in its own statement.)

Here is the part that lit the fuse: the letter "did not provide specific details of its national security concern." The one technical detail in the public record is a jailbreak the government demonstrated, which amounted to asking the model to read a codebase and fix any flaws in it. Anthropic's own response is that this turns up minor, commonplace vulnerabilities, the same ones widely available from other models, and that recalling a product used by hundreds of millions over it would "essentially halt all new model deployments."

So: a flagship model pulled, no public reason given, and the maker openly disagreeing with the call. When a vacuum like that opens up, it fills with theories.

The theories doing the rounds

I am not going to pretend to know which is right. Nobody outside the room does. But the credible ones are worth understanding, because each says something a little different about the world your business is now operating in.

Pick whichever you find most convincing. It honestly does not change what you should do next.

Why the right theory does not matter to you

Because the lesson is the same under all of them. The single most capable tool on the market disappeared overnight, for reasons its own maker did not choose and cannot fully explain, and not one customer could do a thing about it. If you had spent this week wiring something important straight into Fable 5, you would have come in on Saturday to find it simply gone.

That is the whole point, and it has very little to do with Washington. You do not own these models. You rent access to them, and the terms can move under you without warning: the price, the availability, now even the law.

The lesson is resilience, not picking a winner

None of this is a reason to stay out. The capability is real, and as I said last week, it has never been cheaper to get at. It is a reason to use it in a particular way: so that no single model is load-bearing.

That means an operating layer that sits between your business and whichever model is behind it, so you can swap one for another without rebuilding anything. It means your data, your workflows and your governance staying under your control, not welded to one vendor's API. It means treating "which model" as a decision you can change on a Tuesday, not a foundation you pour once. That is exactly what we build and run as M/OS, and it is why we are deliberately fluent across many models rather than wedded to one. When the platform shifts underneath you, and it will, keeping your obligations intact is the work that quietly pays off.

Last week's advice still stands. Go and use this stuff. It is remarkable, and it is cheap. Just build so that losing any one model is an inconvenience, not a crisis.

It may well come back

Anthropic says it is working to restore access, and Fable 5 may be back by the time you read this. But notice that the lesson does not depend on it. Whether the model returns next week or never, the thing worth remembering is that it could vanish at all, with no notice and no say from the people relying on it.

The businesses that do well out of AI this year will not be the ones who backed the cleverest model. They will be the ones who built so it does not much matter which model they happen to be holding when the music stops.