The US government just banned Claude's smartest AI.

Two models. Gone in one evening.

Here's the timeline, in order, so you can feel how fast this moved.

The directive didn't just block one country. It suspended access for any foreign national — inside or outside the United States — including Anthropic's own foreign-national employees.

So Anthropic had no choice. It pulled the plug the same night.

Here's the line that went out to Amazon's cloud, word for word.

What it looks like if you were using it

So if you use Claude every day for normal work — writing, coding, research — nothing changed for you. The ban only hit the two newest, most powerful models. The rest of the Claude family is untouched.

The reason is wild when you hear it.

The government cited national security. But it didn't say much more than that. Anthropic's own words —

So what was the actual trigger?

A company found a way to bypass Fable 5's safety guardrails. The technique was simple. You ask the model to read a specific codebase and fix any software flaws it finds.

That's it. That's the "jailbreak."

And because Mythos is so good at security work, when it goes looking for flaws, it finds real ones. Some of them serious. The fear is that the wrong person could use that to find and weaponise bugs in real software.

But here's the part Anthropic pushed back on hard. The bugs the model surfaced in the demo were nothing new.

In other words — Anthropic says you can do the same thing with OpenAI's GPT-5.5 right now. The capability isn't unique to Mythos. So banning one model doesn't close the door. It just closes one door out of many.

Anthropic complied — but publicly disagreed.

This is the part most people missed. Anthropic did what it was told. Then it put out a statement saying it thinks the whole thing is a mistake.

Their bigger point is about the precedent. If finding one narrow jailbreak is enough to pull a model from the market, then no AI company could ever ship anything. Because every model has edge cases.

And they made one more point that stuck with me. Nobody has shown the jailbreak actually caused harm. It was a demonstration, not an attack.

Anthropic says it believes the government is acting on a misunderstanding, and that it's working to get access restored.

One reporter who covers this space framed the deeper issue better than I can.

That's the real story under the story. Nobody fully knows how to govern a model this capable yet. So the first time it got scary, the answer was: switch it off.

Read it yourself. Don't take my word.

Everything in this guide comes straight from Anthropic's own statement and the major outlets that broke the story. Here's where to read all of it first-hand.

Wait — what even is Claude Mythos?

If this is the first you're hearing of it, here's the quick version.

Claude Mythos is a special Anthropic model built for one thing — finding security holes in software. Not chatting. Not writing emails. Finding flaws.

And it's frighteningly good at it.

That last point is the whole problem in one line. Mythos takes a job that used to need a rare, highly-trained expert — and hands it to anyone.

That's why Anthropic kept it locked down from the start. There were two versions.

Project Glasswing wasn't small either. Twelve giants — including AWS, Google, Microsoft, Apple, Cisco and CrowdStrike — teamed up to patch the bugs Mythos found before the model ever went public.

So that's the backdrop. The most capable bug-finder ever built. A jailbreak. A government letter. And by nightfall, both versions switched off.

The lesson isn't about Mythos. It's about you.

Most people watching this don't use Mythos. You'll probably never touch it.

But the ban teaches the single most important rule of working with AI in 2026 — never build your business on one model you don't control.

Here's the difference, side by side.

That's the whole game now. The model is not the moat. The system around the model is the moat.

Do you need to do anything right now?

Honestly? For most people — no.

If you use Claude for everyday work, you weren't using Mythos or Fable 5 anyway. Opus, Sonnet and Haiku are all still live and working normally.

But here's the one-liner to take away from all of this.

The people who figure out AI now — while the rules are still being written and models come and go overnight — are going to be miles ahead when everything settles. Every system you build. Every workflow you set up. Every model you stop depending on. It all compounds.

Today proved the point in one evening. A model that ranked #1 in the world was switched off for everyone in hours. If your business lived on it, your business stopped.

So the move isn't to chase the next model. The move is to build the system that doesn't care which model is up today.

Recap — what you now know.