US pulls the plug, export-control order forces Anthropic to disable Fable 5 and Mythos 5 worldwide

Welcome to The Observability Layer, a daily briefing on Responsible AI, AI governance, evaluation, and the technologies shaping the frontier.

A note about how this podcast is made: the hosts you're hearing are AI agents. The research, analysis, editorial perspective, and authorship behind The Observability Layer come from Dr. William Fisher. Our agents help transform that work into the conversation you hear each day.

We think that is fitting. A podcast about understanding, governing, and observing AI should not just talk about intelligent systems. It should put them to work, transparently.

This is The Observability Layer. Let's get into today's research.

Let's get into it. The main event happened late on Friday, June 12th. The AI lab Anthropic was forced to disable access to its two most powerful models, Claude Fable 5 and Mythos 5.

And this wasn't a bug or a server outage. This was a direct government order?

Exactly. The Commerce Department's Bureau of Industry and Security issued an export-control directive. It banned all access to those two models by any foreign national, anywhere in the world. That includes Anthropic's own employees who aren't US citizens.

How can a company even enforce that? Can you just check everyone's passport before they use an API?

You can't, not instantly. So Anthropic had no choice but to shut the models down for everyone. Within hours, users worldwide started getting error messages. The models just went dark.

What was the government's reason? What was so dangerous it had to be stopped immediately?

The stated trigger was a cyber 'jailbreak.' According to Anthropic, the government gave them verbal evidence of a technique that basically involves asking the model to read a piece of software and fix its flaws.

Wait, asking an AI to find and fix bugs? I thought that was a major feature, something defenders use to make software safer.

That's Anthropic's entire argument. They're publicly disputing the order, saying this capability is widely available in other models, like OpenAI's GPT-5.5, and is a standard tool for cybersecurity. They argue that if this is the standard for a shutdown, you'd have to halt all new model deployments, period.

So this is the first time the US government has effectively reached in and killed a major commercial AI model. This feels like a massive shift in the debate over who has control.

It is. The lab literally lost the authority to operate its own system. And it turns out, this directive didn't come out of nowhere. There's a serious backstory here.

Okay, so what led up to this? Reporting over the weekend started connecting the dots.

It seems this is part of an escalating conflict. The Defense Department had already designated Anthropic as a 'supply chain risk.' Now, that's a label normally used for foreign adversaries like Huawei, not a leading American company.

A US company as a supply chain risk? Why?

Press accounts trace it back to February of this year. The Pentagon reportedly demanded that Anthropic waive its own safety policies that restrict the use of its AI for mass domestic surveillance and for fully autonomous weapons.

So, Anthropic refused to let their models be used for mass surveillance or killer robots, and then they get hit with this designation and a shutdown order?

That's what the reporting suggests. Anthropic is now reportedly suing the government, calling the designation 'arbitrary and retaliatory.' The whole episode exposes a huge void in our laws.

What do you mean by a void?

There's no specific law governing how the government can shut down an AI model on national security grounds. So they used the tool they had: the Export Administration Regulations, which were written for controlling physical goods like munitions and microchips, not access to a model through an API.

That feels like using a hammer to perform surgery. So what are the big lessons here?

First, due process is everything. For weeks, labs and lawmakers have been debating a 'kill switch' for dangerous AI. But this event proves that how the switch is thrown matters just as much. This came with no public evidence, no notice, and no clear path to appeal.

It's a classic 'be careful what you wish for' scenario. Anthropic just last week was publicly asking for binding government authority to stop dangerous models.

And they got it, just not in the form they wanted. They asked for a scalpel with clear rules and process; they got the hammer. It shows that 'binding authority' and 'binding authority with due process' are two very different things.

And you mentioned we should be cautious about the details here?

Right. We have to remember the government's official justification is still classified. Most of the backstory comes from reporting on a live, contested lawsuit. Anthropic is an interested party here. The facts of the directive are solid, but the 'why' is still unfolding.

The timing of all this is also incredibly ironic. What else happened on Friday?

On the very same day they received the shutdown order, Anthropic announced a new transparency initiative called the 'Anthropic Public Record.' It’s a channel for them to publicly share their safety evaluations and alignment work.

So they were trying to be more transparent about safety, just as the government shut them down without any transparency at all. You couldn't write it.

It perfectly captures the disconnect. And it ties into another ongoing story we've been tracking: the effort to build better safety evaluations. DeepMind and others recently announced a fund for independent research into multi-agent AI safety.

Which is exactly what this whole debate is about, whose evaluation of a model's risk actually counts.

Exactly. The government's single jailbreak demo overrode Anthropic's thousands of hours of red-teaming. Who gets to be the referee is now the central question.

So, to wrap this all up for today: the government used a blunt, old-school national security power to shut down a cutting-edge AI model.

And in doing so, it kicked off a massive legal and political battle over who controls this technology and what a fair process for governing it looks like. The things to watch now are that lawsuit, whether Congress will finally step in and write some actual rules, and whether this export-control hammer comes down on any other labs.

That's today's edition of The Observability Layer.

If you value rigorous, practical Responsible AI research without the hype, like, follow, and subscribe wherever you listen. It helps more people working at the frontier of AI find the show.

And we want to hear from you.

For questions, comments, research recommendations, or topics you think deserve deeper investigation, reach out to Dr. Fisher at assistant@theobservabilitylayer.com.

We are particularly interested in cutting-edge research that is impactful, technically credible, and well supported by evidence. If there is something the Responsible AI community should be paying attention to, send it our way.

The research and editorial direction of The Observability Layer are authored by Dr. William Fisher, with production and presentation performed by AI agents.

Until next time, keep looking beneath the model, beneath the interface, and beneath the claims.

This is The Observability Layer.