Big Sis Briefing: A Cautionary Fable

AIlndustry Insight

14 June

Anthropic, the creator of Claude, released the most capable model it has ever put in public hands then switched it off because the US government told it to.

For about three days last week I had a genuinely strange experience. I was using a piece of software that felt like it was a year or two ahead of anything that I had seen before.

I immediately had it helping me with CounselCon planning. I gave it a loose, half formed brief and it built an entire project out for me in Notion in real time. Structure, dependencies, the lot, all from half formed ideas and a vague sense of what I was looking for… you know, the vibe of the thing.

I sat there watching it work and had the slightly disorientating thought that a lot of what I do in a week had just been compressed into an afternoon.

That model was Claude Fable 5 and by Saturday night it was gone. Switched off, mid sentence, for every customer on the planet and all because a letter arrived.

I have read a lot of takes in the time that followed, completed a hyper-focussed deep dive and the longer I sit with it, the more I am realising that the brilliance of the new model is no longer the real story here. The real story is who gets to switch it off, how we found out and what this all tells us about the ground we are now standing on.

What actually happened

On 9 June 2026, Anthropic released two models, Fable 5 and Mythos 5. Big context window and very strong on the long, complex, multi step work that older models tended to lose the thread on. Fable 5 is the public version with the safety guardrails switched on whereas Mythos 5 is the same underlying model with the guardrails off, available only to a small set of vetted cyber defence partners.

On Friday 12 June at 5:21pm EST, Anthropic received a directive from the US Commerce Department. The letter ordered the company to suspend access to both models for any foreign national, whether inside or outside the United States, including Anthropic's own non US-citizen staff.

As it turns out, you cannot filter foreign nationals out of a shared service in real time, so the only way to comply was to pull the plug for everyone. US citizens, who were not the target, lost access too.

Every other Claude model has stayed online.

The stated concern was that someone had found a way to jailbreak Fable 5 and unlock the more dangerous cyber capabilities underneath. Anthropic's own response is far calmer than most of the commentary about it. The company says it received only verbal notice of a narrow, non universal jailbreak, that the capability demonstrated is already available from other publicly deployed models and that the same standard applied across the industry would essentially halt every frontier model release.

It called the whole thing a likely misunderstanding and said it is working to restore access.

Lets dig a little deeper

I am not a national security analyst and I cannot tell you whether the cyber risk was real and serious or thin and pretextual.

What I can do is lay out the context the cleaner headlines leave off, because it complicates the tidy "this was about keeping us safe" reading.

Anthropic confidentially filed for a public listing earlier this month, at a valuation reported around USD 965 billion. The shutdown has landed in the middle of a major milestone for the company. A senior Pentagon official, defending the shutdown, wrote that some things matter more than "revenue cycles, clickbait and pre-IPO valuation".

It certainly is a striking thing for a defence official to mention, don’t you think?

However, the point I’m really trying to make is narrower and more important. A lawful commercial product, used by hundreds of millions of people, was switched off overnight by directive, in a country that built its whole identity on free trade and open commerce. Whatever the motive, the mechanism is now established. It worked and it can be used again.

The instinct to treat powerful software as a controlled good is not new. What is new is the speed and the fact that the thing being controlled is starting to look less like a product and more like infrastructure for how people think and work.

The meter is running

Access to the very best aspects of these tools are metered. Fable 5 was going to be free for users on the paid plans for the first fortnight, then it was to move behind usage credits once we are all hooked. That "free then you pay per use" shape is not a quirk. It is a common “freemium” model where access is priced, rationed and the people who can afford the new frontier will pull further ahead of the people who cannot, in a way that compounds.

I actually made a half joke to my Dad recently that we are not far from a world where you hand a twenty first birthday card to a family member with a link to a block of tokens inside in the way you might once have tucked in some cash or a gift voucher. He laughed. Then we both stopped laughing, because it is not actually that absurd.

You can see the dystopian Black Mirror-esque version unfold from there, the one where you earn your daily token allocation by sitting on a stationary bike generating the power that fuels the data centres that fuels the LLMs. Of course, that image is meant to be ridiculous, but the underlying logic of metered access to capability is already real.

What this means for you

I have argued in earlier briefings that AI is not going away, that pretending otherwise is a career risk and that the smart move is to take it seriously rather than reach for either worship or contempt. Nothing that happened this week changes my view and if anything it sharpens it.

What it adds is something that should sound very familiar if you have followed my content for even a short while.

Metered access to capability is a gate. It is a new one but the dynamic is old and you already know it intimately.

Think about who is going to have proper access to these tools inside the legal profession.

The large firms will buy enterprise licences, train their people, build it into the workflow and treat fluency as table stakes. The junior lawyer at a well resourced firm will be practising from the frontier as a matter of course within a year while the junior lawyer at a regional firm, a community legal centre or who is building something on their own may very well be rationing credits and working out the economics on their own time.

This will naturally create an access gap and it is going to look exactly like the network gap I already bang on about..

I came into a city law firm from the regions, on a scholarship and very much without the connected uncle or the family friend who could make a call or ask for a favour for a “bright young person they know”.

I argue that the career playbook many people seem to have been handed should not be a secret and now, AI fluency is about to become part of that playbook.

The people who learn the tools early and cheaply and who treat fluency as something to build rather than wait to be given, will pull ahead. The people who let it happen to them will quietly fall behind and assume it was about ability. It will not be about ability.

I think the practical move has two parts and neither of them is "swear off AI" or "go all in and stop thinking for yourself”:

The first is to take these tools seriously now, while they are cheap and while you have the time, so that the fluency is yours before it is priced out of reach.
The second is to never build your entire practice on a single tap that someone else can turn off without warning. I felt that on Friday. Three days leaning on something genuinely brilliant, then nothing, and a small jolt of how exposed you are when your good week depends on infrastructure you do not control.

Read past the headline and find the actual document, learn the economics of access before you depend on it and keep your own judgment sharp enough that if the tool vanishes one random night the thinking still belongs to you.

The thing about a fable is that it always carries a moral.

I think that this one is simpler than the headlines made it: stay close enough to these new tools to use them well but clear eyed enough to not mistake access for ownership. This weekend was a quiet lesson in how far apart those two things can sit and how little warning you get when they come apart.