Shut down the experiment, not the Earth.
That is the rule.
If a human-made system becomes dangerous enough to threaten the conditions of life, the duty is to stop the system.
By conditions of life, I do not mean inconvenience, embarrassment or political discomfort.
I mean the basic ground a society cannot casually trade away: life, safety, democratic agency and ecological conditions that may not be recoverable once lost.
Do not sacrifice the world to the experiment.
Do not treat collapse as destiny.
Do not confuse technical momentum with moral permission.
And do not let the people building the experiment describe stopping as the only unthinkable option.
This is not a call for panic.
It is a call for a sane hierarchy.
The world comes first.
The experiment comes second.
Modern systems often behave as if that order is controversial.
Once enough money, status, infrastructure, talent, institutions, markets and national competition gather around a technical project, stopping can begin to sound impossible even before the risk is understood.
The investment is too large, the race too advanced, and the institution already reorganized around continuation.
Then the question changes quietly.
Instead of asking whether the experiment should continue, society starts asking how to survive the fact that it is continuing.
That is the wrong question when the harm becomes large enough.
No experiment has a right to become unavoidable.
No technology has a right to turn its own momentum into a planetary claim.
No institution has a right to say the world must adapt because stopping would be inconvenient.
AI makes this rule urgent because it is not only a tool category.
It can become a decision layer, a persuasion layer, an administrative layer, a market layer and an infrastructure layer.
That does not make AI evil.
It makes limits on its role essential.
The correct question is whether this can be stopped, slowed, isolated, rolled back, refused or redesigned before the harm becomes normal.
That question should exist before the emergency.
A serious society does not wait for collapse, injury or poisoned rivers before asking whether inspection, trial rules or suspension powers should exist.
Nor should it wait until an AI system becomes structurally dangerous before asking whether the experiment can be halted.
The ability to stop is part of responsible design.
An experiment without a stop condition is not mature.
It is faith wearing technical language.
Governance often fails by imagining rules as something added after progress, brakes as enemies of innovation and caution as weakness.
But brakes are not the opposite of movement. They are one of the reasons movement can be trusted.
A vehicle without brakes is not more advanced because it goes faster.
It is less fit for public life.
If a system cannot be paused, audited, separated, corrected, exited or shut down, then it has been built with a political assumption hidden inside it:
continuation above accountability.
But stopping power is also dangerous.
The threshold cannot be whatever a frightened government, powerful company or political majority declares it to be.
A halt must answer to evidence, proportionality, independent review, and the scale, likelihood and reversibility of the threatened harm.
The power to halt a system must itself be limited, reviewable and accountable.
Otherwise the brake becomes another experiment without brakes.
The duty to halt an experiment is not hatred of science, rejection of tools or refusal of the future.
It is one of the conditions for science to remain civilized, tools to remain tools and one path not to consume all futures.
This matters beyond AI.
Societies build experiments all the time: financial systems, surveillance systems, agricultural systems, labor systems, platform systems, weapons systems and administrative systems.
Some become useful. Some become harmful. Some begin useful and become harmful when scale, incentives or capture change them.
The mature question is not whether experimentation should exist. Of course it should.
The mature question is whether society keeps enough authority to halt an experiment when it starts damaging the conditions that justify it.
Innovation, markets, institutions, models and races are not above life.
When a system becomes harmful enough, the duty is not to decorate the harm with better language or optimize the damage.
The duty is to halt the experiment.
Carefully.
Publicly.
Accountably.
With repair paths, evidence, oversight and protection against abuse.
But still halt it.
The world is not a test environment for every system powerful people want to continue.
If the experiment threatens the conditions of life, shut down the experiment.
Not the Earth.
Source Boundary
This piece is a public-note translation of an internal AI-governance and system-containment claim. It does not call for sabotage, violence, panic, anti-science politics, or harm toward people, Earth or life. It does not claim that all AI systems are existentially dangerous or that any specific shutdown policy is already proven. Any later empirical claims about AI risk, safety governance, emergency powers, infrastructure separation, model recall, shutdown protocols or technology moratoria should be source-checked separately.
