AI is moving too fast.
We need a
speed limit.

Everyday people agree

People from all walks of life agree on slowing down AI. Everyday people, AI researchers, and Nobel laureates have all called for a slowdown. But AI corporations are still racing recklessly.

of every 100

72 want AI slowed down8 want it sped up20 not sure

National poll of 1,001 US voters, asked whether AI development should be slowed down or sped up. Source ↗

71%

of Americans say AI is moving too fast

A strong majority of Democrats and Republicans. ↗

1,386

AI company employees have asked the government to slow down

Two “godfathers of AI” agree. ↗

People sitting on a wall in silhouette against a dusk sky

No one asked for, or approved, the AI now being built.

And now, AI is putting our loved ones, our communities, our jobs, and our world at risk. The government must step in to hold AI corporations accountable and stop them from making these dangers worse.

We Must
Slow Down AI

That means the government limiting the development of the most powerful AI systems and limiting AI from entering our jobs, our kids' schools, and our communities. AI corporations must not dictate the terms of a slowdown.

Ways to slow down AI:

01Enforce a speed limit on the development of the most powerful AI systems.

We demand that the government set enforceable limits on how quickly AI corporations can advance the capabilities of their most powerful systems. No company should be allowed to race ahead faster than society can understand and prepare for what it's building.

02Ban superintelligence.

The development of superintelligent AI, which we take to be AI that is smarter than all humans combined, would pose unacceptable risks to humanity.

We believe governments should prevent superintelligence from being built without public buy-in and scientific consensus that it would be done safely.

03Negotiate an international slowdown.

We demand that the U.S. government pursue a verifiable agreement with China and other leading AI nations to slow the development of the most powerful systems. The agreement must establish shared speed limits, independent inspection requirements, and clear penalties for violations so no country can gain an advantage by secretly racing ahead.

04Slow down AI-driven job displacement.

Advanced AI is capable of automating human jobs rapidly, and on a large scale. Large-scale job loss could destabilize communities and fuel social upheaval. The government must act now to prevent this risk and ensure society has time to adapt.

05Slow down AI’s rollout to children.

We demand the government slow AI’s push into children’s classrooms and emotional lives. Our precaution must extend beyond the classroom: AI companions and therapy bots must remain adults-only until independently proven safe for minors.

06Slow down the AI race by making AI companies, and their executives, liable for the harm they cause.

We demand that individuals and communities harmed by AI must be able to seek full compensation from the companies that build or deploy it. Executives who knowingly cut corners on safety or deceive regulators should face personal legal consequences.

We need to slow down AI on every front.

Common
Questions

Is a slowdown possible?

Yes. We have imposed limits on other dangerous technology before: chemical and biological weapons, nuclear weapons, human cloning.

Each of those technologies had powerful actors pushing for unrestricted development, and in each case, common sense prevailed. Further, many of these efforts required and were successful in winning international coordination.

Scientists have successfully slowed their own field, too. Molecular biologists paused recombinant DNA experiments in 1974 and only resumed after agreeing on critical rules.

In a similar fashion, AI can also be slowed down.

Who benefits from a slowdown?

The public benefits. A slowdown hurts big AI corporations the most, while giving everyone else more time and power to shape how AI enters their workplaces, communities, and lives.

Who sets the terms of a slowdown?

Governments must set the legally binding terms of a slowdown. AI corporations must not dictate the terms of a slowdown.

What about China?

China's smartest AI systems come from copying US AI. If the US slows down, China slows down.

Next, the US can propose to cooperate and make this more robust with onsite inspectors. This is incentive-compatible for China. If both sides are slowed by roughly the same factor, then this does not really harm China’s competitiveness but it makes them safer because the risk of losing control of AI is reduced.

Then, if China refuses to slowdown and safeguard their future AIs, there are a variety of measures the US could take. To get an agreement, the US can apply typical forms of negotiation pressure. The US could also disrupt their AIs and interfere with their AI development or deployment.

The US should begin with unilateral restraint, then propose to cooperate. If that fails, it can consider measures to counteract them.

Working alongside