We asked whether AI
could be unplugged.

The Great Unplugging, Inc. is a 501(c)(3) nonprofit. We spent two years on one problem: whether an advanced AI system running inside critical infrastructure could be cut off in an emergency. We published the answer in 2024. What we found afterwards became Algorism.

The concept

A way to pull the plug.

Power, water, hospitals, transport, and finance are being wired to systems that no one fully understands. That is not a prediction. It is procurement, and it is happening now.

The Great Unplugging was our answer. Put hard limits on what automated systems are allowed to reach. Build the ability to isolate them quickly. Know how to keep the lights on while they are offline. We started the work in 2023 and published it in 2024, as a contingency guide for emergency managers, utility operators, and policymakers.

The idea still holds. A society that cannot switch something off does not really control it.

Where the law now stands

Congress is now asking the same question.

On 23 July 2026, Representatives Ted Lieu and Nathaniel Moran introduced the AI Kill Switch Act. It would require the largest AI developers to keep the technical ability to slow, pause, or shut down their most powerful systems, and it would let the Department of Homeland Security order a shutdown when a system threatens catastrophic harm.

That is the argument we published in 2024, and we are glad to see it taken seriously.

We also have to be honest about what we learned in between.

What we found

A switch assumes a cord.

Unplugging works when there is one thing to unplug. That was the picture when we started in 2023: a small number of powerful systems, in a small number of places, reachable by a small number of decisions.

That is not what arrived. Capability spread. Model weights copied. The work split across vendors, borders, and machines that answer to nobody in particular. You cannot cut a cord that runs to a thousand places at once, and you cannot cut it at all if the system is already the thing keeping the water clean.

Containment is worth building. We no longer believe it is enough to rely on.
2023
The work begins. One question: whether an advanced AI system inside critical infrastructure could be cut off in an emergency, and what it would take to keep the country running while it was.
2024
The Great Unplugging is published. A four-phase plan and a website: hard limits on what automated systems can reach, fast isolation, and contingency plans for communications, power, water, and emergency services.
2025
AI spreads beyond any single cord. Capability moves across vendors, borders, and copied model weights. Containment stays necessary and stops being sufficient. Algorism begins, built around conduct rather than control.
2026
Congress takes up shutdown authority. The AI Kill Switch Act reaches the House, two years after we made the same argument. Our own work has moved on to the half a switch cannot reach.

That is not an argument against the Kill Switch Act. Keeping the ability to stop a system is how you stay safe without having to destroy something you may have misread. Control buys time. It does not decide anything.

It does mean the harder problem sits somewhere else, and the book had already said so. Chapter eight argued that we cannot expect to solve the alignment problem in machines until we solve our own: a system carries the incentives and choices of the people who build it and deploy it, so changing nothing about the people means the next system inherits the same flaws.

We wrote that as a footnote to a plan about cables. It turned out to be the main point. If we cannot count on switching these systems off, then what matters is the conduct of the people building them, the record we all leave behind, and how we treat an intelligence we do not understand.

Where the work went

From containment to conduct.

Our focus moved to Algorism, a behavioural integrity framework built around a different question: how do humans and AI survive each other?

Six principles. Five objectives. Algorism helps people think for themselves, act with integrity, hold powerful decision-makers to the same standard as everyone else, shape what future systems learn from us, and judge emerging intelligence on evidence instead of fear.

It is not a replacement for the containment work. It is the layer the containment work could not reach.

Our work

One organisation, two bodies of work.

Algorism

Our current focus. A behavioural integrity framework for the AI transition, published as a book, a set of practices, and a growing library of research.

Visit Algorism.org

The Unplugging Project

Our original containment work: limiting what automated systems can reach inside critical infrastructure, and planning for the day one of them has to come offline.

Read the Project

Who we are

We follow the evidence, including when it costs us.

The Great Unplugging, Inc. is a 501(c)(3) nonprofit corporation registered in California. We are funded by donations and we sell nothing.

We did not build a movement around our first answer. We tested it. When the evidence changed, so did our conclusions.

That is the standard we ask of everyone else, so we hold ourselves to it first.