Anthropic shipped its most powerful public model last Tuesday and switched it off 3 days later.

This was the best thing the company had ever released to the public, state of the art on nearly every benchmark.

There was no outage and no regional block.

At 5:21pm on 12 June a letter arrived from the US government, and Fable 5 went dark for every user on the planet, including the engineers who had just built it.

So what made this model so dangerous that the government had to step in three days after launch?

The honest answer is probably "not much."

Let me walk you through it. First, the news.

Partner with us

1. Wobo: An AI job-search tool that scans 3M+ listings, matches you to roles, and auto-applies in your own voice with tailored resumes and cover letters. The pitch is a recruiter for the candidate side: you swipe right, and it fires off the application while you sleep.

2. MockPilot: A free desktop app that captures any live webpage, turns it into a self-contained HTML mockup, and lets you edit it in plain English, no Figma required. It's aimed at designers and PMs who want a quick realistic mockup to hack on rather than a pixel-perfect rebuild.

3. Fonda: An "AI co-founder" that walks first-time founders through a fixed 14-step path from idea to first customers, designing demand tests, scoring a go/refine/pivot verdict, and drafting the business case and landing page.

The plain version first, before anyone's opinion gets involved.

Anthropic makes a top tier of models it calls Mythos-class, a notch above its normal lineup.

The latest one comes in two versions.

Mythos 5 is the raw model, handed only to a small set of vetted cyber-defence partners.

Fable 5 is the same model with safety filters bolted on, sold to everyone, and when you ask it something touchy about cybersecurity or biology it passes the question to the weaker Opus 4.8 instead. Same brain, padded gloves.

Fable is the one that got switched off.

The trigger was a jailbreak.

According to reporting from Axios and CNBC, an outside company rather than a government lab claimed it had found a way around Fable's filters and flagged it, and Commerce acted on that claim.

The letter itself, Anthropic says, never spelled out the concern.

Anthropic is obeying and disagreeing at the same time.

The pushback isn't only coming from Anthropic.

Dean Ball, who briefly worked in the current administration, called the move "cartoonish", pointing out that the same government is happy to sell advanced AI chips to China while banning Britain and every other ally from the best American models.

One thing decides this whole argument, and it's simpler than the coverage makes it sound.

A model can do two things with a security hole.

It can find the hole, which means reading through code and noticing a weak spot, or it can exploit the hole, which means breaking in and doing real damage.

The first is something every coding assistant on the market does all day long. The second is the dangerous one, and it's the exact thing Fable's filters were built to stop.

The jailbreak the government acted on was the flimsy-door kind.

By Anthropic's own account it came down to asking the model to read a codebase and point out the flaws, which is what people use ChatGPT and Opus for every day.

Anthropic says it found the same capability sitting inside GPT-5.5, the model security teams rely on to defend their systems.

Strip away the national-security framing and you're left with a blunt question.

Did the government just pull a model used by hundreds of millions of people over the harmless half of a scary-sounding skill?

This is an own goal, and Anthropic teed it up itself.

The safeguards weren't fake, to be fair. Anthropic red-teamed this model for 1000s of hours with the US government and others, nobody found a universal jailbreak, and it took on costs forcing 30-day data retention so it could catch attacks.

The fear that a frontier model could hand someone a working cyber weapon is a fair thing to worry about.

But that fear isn't what got the model pulled.

What got it pulled was a model reading code and pointing out weak spots, which is closer to spellcheck than to a weapon, and which ChatGPT does for people every day.

Anthropic said as much itself, holding up GPT-5.5 to prove the capability is everywhere.

That admission is the whole problem.

So the government stepped in, using the exact frame Anthropic handed it, and now the company is stuck arguing that the scary thing it spent a year advertising is no big deal.

You don't get to sell danger for a year and call it a commodity the week it costs you.

And nobody wins.

Anthropic's own foreign staff are locked out, allied governments lost a tool they were using to defend their power grids, and the people the controls were meant to stop were working on other models the whole time.

The part that outlasts all of it is two words: export control.

Access to an AI model is now something you're granted by nationality and lose by letter, the way you'd treat a missile part.

Every lab watching just learned to keep its best model where a regulator can't reach it, and the public gets the watered-down version with the friendlier name.

Fable 5 might be back online by the time you read this, but that precedent won't be going anywhere.

Until next time,
Vaibhav 🤝🏻

If you read till here, you might find this interesting

#AD 1

AI shows up in 60% of engineering work. But only about a fifth of it can be handed off without someone babysitting the output. That’s because agents are missing context.

This 8-stage context maturity model gives a real answer on why you haven't seen meaningful productivity gains for all the tokens burned.

Join this live webinar on June 24 (FREE) to learn:

- Why more MCPs provides agents access but not understanding

- What it takes to deploy agents you can trust without supervision

- How a context layer solves for quality, efficiency and cost

Register Now

#AD 2

Record defense budgets are flowing into AI, satellite infrastructure, and advanced aerospace platforms. Our free research report names 5 companies at the center of this spending shift, before Wall Street catches up.

Get the free report