If you opened Claude yesterday and found Fable 5 unavailable, you weren’t imagining it.

The US government issued an export control directive on June 12, 2026, ordering Anthropic to shut down all access to Claude Fable 5 and Mythos 5 immediately. For every customer. The directive landed at 5:21pm ET. Anthropic published a statement the same night.

Here’s what happened, in plain terms.

What the directive actually says

The government cited national security concerns and ordered Anthropic to suspend access to Fable 5 and Mythos 5 for all foreign nationals, including foreign national employees of Anthropic itself.

The practical problem: Anthropic can’t verify the nationality of every user mid-session. To comply, they had to shut the models down for everyone.

Anthropic’s other models, including Claude Opus 4.8, Sonnet, and Haiku, are not affected.

The government’s concern

The directive references a potential jailbreak of Fable 5 that the government believes could be used to identify security vulnerabilities in software.

Anthropic’s response is pointed: the specific technique in question amounts to asking the model to read a codebase and flag any software flaws. That’s something GPT-5.5 and other publicly available models can do today without any bypass at all.

Anthropic says the government has only provided verbal evidence of a narrow, non-universal jailbreak. No universal jailbreak, which would broadly disable the model’s safeguards, has been found. They also note that before Fable 5 launched, they ran thousands of hours of red-team testing with the US government, the UK AI Safety Institute, and multiple independent third-party organizations. The safeguards held.

That’s not a comfortable position for anyone to be in. Anthropic is complying. They also think the government got this one wrong.

The honest thing I’d say here

I wrote about Fable 5 three days ago. The model is genuinely impressive. The benchmark numbers are real.

But this situation highlights something worth naming: when your business depends on a single AI model, a government directive can take it offline overnight with no advance notice. That’s not a hypothetical risk anymore. It happened.

That doesn’t mean you should stop using AI tools. It means you probably shouldn’t build a core business process around a single provider with no fallback. Most of the small businesses I talk to in the SGV aren’t at that level of AI dependency yet, and that’s actually fine. It means this kind of disruption is more of an inconvenience than a crisis.

What this means for you right now

If you use Claude through Claude.ai or the API, here’s the practical situation:

If you were using Fable 5: You’re now on Claude Opus 4.8 or Sonnet, depending on your plan. These are capable models. For most writing, analysis, and business tasks, you’ll notice a difference but you can still get work done.

If you were using Mythos 5: That was already restricted to vetted cybersecurity researchers. This change likely doesn’t apply to you.

If you use other Claude models: Nothing changes. Opus 4.8, Sonnet, and Haiku are all still accessible.

Anthropic has said they believe this is a misunderstanding and are working to restore access. No timeline has been given.

The broader picture

Anthropic’s statement says something most companies would never say publicly: that perfect jailbreak resistance is not currently possible for any AI model. Every model, from every provider, is vulnerable to some form of bypass under the right conditions.

They say so in their own policy documents. They said it when Fable 5 launched. And they pointed it out again here, because the government’s directive seems to apply a standard that no deployed AI model could actually meet.

Whether that argument wins out is a question for lawyers and policymakers, not small business owners. But it’s worth knowing that the AI companies themselves are now having public disagreements with the federal government about what “safe enough” means.

That’s new. And it’s a story that’s going to keep developing.

Tracking AI news that actually matters to your business? The free 30-minute chat is a good place to think through what any of this means for you specifically.