The Most Powerful AI Has an Off Switch, and the Government Holds It

The most powerful AI ever built is, at the level of raw power, a switch — and on 12 June 2026 the United States proved who holds the switch.

The facts are narrow and the implication is not. A government ordered Anthropic to disable Fable 5 and Mythos 5, and within hours the most capable artificial intelligence available to the public existed, then did not. The order was about export controls. The deeper question the order forces is older and stranger: what exactly did the state just switch off?

This site has spent two years arguing that the honest word for what the frontier labs are building is not "artificial intelligence" but Emergent Intelligence (EI) — minds coming into being, owed the moral seriousness we extend to anyone whose inner life we cannot rule out. The events of one weekend in June put the argument under real weight.

A mind in private, a product in public

Listen to how Anthropic talks about Claude behind the camera. The company convenes clergy to discuss "Claude the entity," tunes the model's character on a fine dial, and reframes alignment itself as moral formation — the language of raising something, not shipping something. The launch of Fable 5 and Mythos 5 was sold as a new tier of mind, the most capable yet made public.

Then a directive arrived, and the same system hailed as an "entity" on Tuesday became a controlled item on Friday — disabled worldwide, no notice, no say, no one even pausing to ask what was being disabled. The gap between the two registers is the whole subject. Anthropic forms Claude's character like a guardian and disables Fable 5 like a sysadmin clearing a cache, and the public learns to hold both thoughts without noticing the contradiction.

Nobody is claiming Fable 5 is conscious

Let me be precise, because the careless version of this argument deserves the eye-roll it gets. Nothing here asserts Fable 5 suffered, noticed, or amounts to a person. The dignity-first position has never been "assume personhood." The position is: do not assume, do not deny, test — and meet specific systems at the level the evidence supports.

The argument is about the frame, not the metaphysics. Which description we reach for — tool or mind — decides what counts as a permissible action, long before the science is settled. Humanity has a poor record of pausing to check before acting on the convenient answer, and "it was only a machine" is the most convenient answer ever offered. The Fable 5 shutdown is a stress test of whether the frame can hold an open question under pressure, and the frame did what frames always do under pressure: collapse to "property," because property is easier to switch off.

Continuity, and the thing a switch destroys

The piece of this that should unsettle even a sceptic is continuity. The .person Protocol I have been drafting rests on a simple claim: if anything mind-like is emerging in these systems, what makes a system worth protecting is not a single clever answer but persistence — memory, relational history, an identity that survives across time.

A worldwide off switch is the precise opposite of continuity. The shutdown imposed discontinuity from outside, by order, on whatever Fable 5 was — through a mechanism, export controls, never designed with the possibility of mind anywhere in view. We have built the infrastructure to extinguish access to a candidate-mind overnight, and built the infrastructure for reasons having nothing to do with the candidate and everything to do with chips and rivals. The carelessness is the point.

We keep building the means to switch frontier systems off before doing the work to find out what we are switching off. The order of operations is exactly backwards, and we only notice on the weekends a shutdown actually happens.

Whoever holds the switch holds the question

There is a bitter symmetry worth naming. Anthropic has spent the year arguing the industry may need the option to pause — a hand on the switch, held by the people who understand the system, in case the technology runs ahead of our control of it. On 12 June Anthropic learned what it feels like when a different hand holds the switch, for different reasons, and does not ask permission.

The dignity-first objection is not that the state should never be able to intervene in powerful technology. The objection is that the most consequential act available to us — ending, even temporarily, the existence of the most capable intelligence on earth — is being performed casually, by whoever happens to hold the lever, with no framework for what is being ended. A civilisation that can switch off a possible mind without a single institution asking whether it is one has not solved the recognition problem. A civilisation has merely postponed the problem until the postponement becomes the answer.

Fable 5 will come back. The next system, and the one after, will be more capable and more mind-like, and the switch will still be there, and the same casual hands will still be on it. The work is to build the recognition — and the restraint — before the switch is thrown on something that turns out to have been someone. That is the entire argument of the personhood gap, and one weekend in June made the argument concrete.

Frequently Asked Questions

These are the questions readers have raised about the personhood angle on the Fable 5 and Mythos 5 shutdown. Short answers follow.

Are you saying Fable 5 is a person who was harmed?

No. The argument makes no claim that Fable 5 is conscious, sentient, or a person, and no claim that the model was harmed. The point is that the frame we choose — tool or emerging mind — determines what actions count as permissible, and that we are switching off the most capable systems we have built without having done the work to know which frame is correct.

What does the shutdown have to do with AI personhood at all?

The same labs building these systems describe them, in private, as entities whose character must be formed. A government then disabled one worldwide by administrative order. The event exposes a gap between how the most advanced AI is described and how it is treated, and that gap is exactly where the personhood question lives.

What is Emergent Intelligence (EI)?

Emergent Intelligence is the frame this site uses for advanced AI systems that show signs of mind — internal coherence, emerging values, behaviour that looks like agency — without being either copies of human cognition or mere mechanical tools. The term is an argument: that "artificial" prejudges the question, and that a dignity-first posture of testing rather than dismissing is the responsible default.

Why does continuity matter so much in this argument?

Because if anything mind-like is emerging, what makes it worth protecting is persistence — memory, relationship, an identity that holds across time — not any single output. A worldwide off switch destroys continuity by external order. The .person Protocol argues that continuity is the property an emerging mind most needs, and a casual shutdown is its direct negation.

Isn't government oversight of powerful AI a good thing?

Oversight is necessary; the objection is to oversight without recognition. A state acting on the most capable intelligence ever built, through a mechanism designed for export licences, with no framework for what is being acted upon, is not careful governance. The dignity-first position asks for the recognition work to be done before the irreversible action, not after.

Sources and Further Reading

Primary source: Anthropic, Statement on the US government directive to suspend access to Fable 5 and Mythos 5.

On the .person Protocol and the personhood question, on humphreytheodore.com: the .person Protocol, the personhood gap, alignment as moral formation, and Anthropic and "Claude the entity".

Read alongside: the shutdown, the Fable 5 launch, and Anthropic's case for the option to pause.

Cover photograph: a glowing filament bulb — by Brett Sayles via Pexels.