Welcome back to the AI For Humans newsletter!

For a brief, beautiful four-day stretch, we (the general, non-enterprise public) got our hands on the newest, most cutting edge AI model.

Claude Fable 5 was, in a word, incredible.

You probably saw our coverage on this week's episodes.

And then, on Friday evening, all of that changed.

The US government, citing national security authorities, ordered Anthropic to suspend all access to Fable 5 and Mythos 5 for any foreign national, inside or outside the country, including Anthropic's own foreign-born employees. 

The only way to comply? Pull the model for everyone.

So how did the most powerful model the public has ever touched vanish in 96 hours? And what happens next?

Let's get into it.

Please support AI For Humans by learning about our sponsors below:

Most AI tools ask you to trade your data for intelligence. Norton Neo doesn't. It's the first safe AI-native browser built by Norton, and it gives you powerful built-in AI without handing your privacy over to get it. Search, summarize, and write with AI built directly into your browser. Your data stays yours. Your context stays private.

Built-in VPN, anti-fingerprinting, and ad blocking come standard. No add-ons. No setup. No compromises.

Fast. Safe. Intelligent. That's Neo.

Download Norton Neo

Okay, quick primer, because there's a AI-centric word at the center of this whole thing you need to understand: jailbreak.

A "jailbreak" is when someone tricks an AI model into doing something its safety training is supposed to prevent.

Think of it like talking your way past a bouncer. The rules are still technically there, you've just found a sentence that gets you through the door anyway. 

Every model has them. It's a constant cat-and-mouse game between the labs and the people poking at their systems.

And this entire kerfuffle with Claude Fable 5 is because of a supposed jailbreak.

Below is a rough timeline of how this all went down:

That is a wild 24 hours.

This isn't a simple story and there’s a lot of stuff going on the background.

There are two sides at play here (at least as far as we know), and how you feel probably depends on how much you trust each player.

The government's side, laid out by AI czar David Sacks: Anthropic spent years telling everyone Mythos was so dangerous it needed government-level regulation. 

So when a "highly credible trusted partner" found a jailbreak that unlocked its cyber capabilities, and the admin asked Dario Amodei to fix it or pull it, he refused. To them, that's a safety company suddenly choosing its consumer product over the safety it built its whole brand on.

Love this newsletter? Forward it to one curious friend. They can join in one click.

Anthropic's side: they reviewed the demo and found it surfaced a "small number of previously known, minor vulnerabilities," stuff other public models, including GPT-5.5, can already find without any bypass at all. 

Their sharpest line: "If this standard was applied across the industry, we believe it would essentially halt all new model deployments." 

Worth knowing this didn't happen in a vacuum. Anthropic and this White House have been circling each other for a while. Sacks has previously accused the company of running "a sophisticated regulatory capture strategy based on fearmongering."

So depending on your read, this is either a real safety incident or the latest round of a long-simmering beef finally boiling over. (There are also reports the export move was tied to suspicions a China-linked group accessed Mythos, which, if true, reframes a lot of this.)

The simplest "why" we know right now? Anthropic was told to fix or pause, said the issue wasn't serious, and didn't move fast enough. As one source put it to Axios: "they were overly confident."

A few things I'm fairly confident about but this is a situation in major flux.

First, I suspect we get Fable 5 back, and reasonably soon.

Even Anthropic has signaled this specific jailbreak capability likely exists in GPT-5.5 too, which makes a permanent, Anthropic-only ban hard to justify. 

The administration itself has said it doesn't view other models as threats yet, only things at "Mythos level or above." This reads more like a pressure tactic than a death sentence.

So no, I don’t think the American government (especially this regime) is willing to pause our entire AI efforts right now.

But is MUCH bigger than one model: this cracks open the AI nationalization conversation like we haven't really seen before.

One source described the export letter to Axios as a "de-facto licensing regime."

If you've read essays like AI 2027 or Leopold Aschenbrenner's Situational Awareness, this is the exact inflection point they keep circling: the moment the state decides these systems are too strategically important to leave fully in private hands.

Which means we're about to find out how good "good" really is.

Look, I don't think this slows down AI progress one bit. However, it might actually slow down when you get access to that AI progress.

For the last couple of years, AI has mostly lived in apps and chat windows and our little creative experiments. 

This is the first time the most powerful version of it got treated like what the labs have been telling us it is the whole time: critical national infrastructure that governments fight over.

This is roughly what "the technology is getting smarter than us" looks like in practice.

Not a dramatic robot uprising, just a quiet 5:21pm letter and your access revoked right before dinner.

We'll be talking about this Friday-night takedown for a long time.

-Gavin

This week on AI For Humans: Fable 5 (when it was available) Was Amazing👇

AI creators are proliferating fast, and the tricky part is that so much of the work starts to look the same.

That is absolutely not the case with Captain HaHa.

He's been locked in on a very specific aesthetic, recreating a weird, hazy '60s/'70s vibe, and recently landed on something close to a sci-fi Sid & Marty Kroft style that's genuinely unlike anything else in my feed.

In a world drowning in same-y AI output, a real point of view matters more than ever. And if that POV is weird 1970s puppets, all the better!

Go check out his work and give him a follow!