To build, or not to build?

Let’s pause superhuman AI development until we all get a say.

A few companies are racing to build AI smarter than every person on earth, in 2027 based on the best predictors’ median estimates. Once it’s built, there’s no going back. Calculate your odds in about a minute to see what you think the future looks like.

Already know where you stand? Skip to take action →

You’re not alone

64%of Americans say AI is developing too fast, and not everyone knows that superhuman AI could arrive in 2027.AP-NORC, Oct 2026
8 in 10favor slowing or stopping powerful AI until its safety risks are better understood.Quinnipiac, Oct 2026
73%are concerned future AI could threaten human survival.Quinnipiac, Oct 2026

Concern crosses party lines. Most Democrats and Republicans agree AI should stay under human control, and the current consensus is that the progress toward the alignment needed to keep humans in control is significantly lagging behind the development toward superhuman AI.

The issue in 60 seconds

The problem

Not the AI you use today. The frontier labs are building a system smarter than all of us combined, able to improve itself without us. Sometimes what we call superhuman AI is referred to as AGI as well. Nobody knows how to keep it under control.

Why now →

The proposal

A verifiable, international and domestic pause before anyone crosses that line, modeled on nuclear arms control. Today’s AI keeps improving. We just don’t build the thing we can’t take back.

What a pause looks like →

Your part

Lawmakers and leaders move when enough people speak up. Enter your address and email everyone who represents you, in about five minutes.

Take action →

The outcome of that choice affects all of humanity. So, the decision should be made by all of humanity, and not just a few technology leaders at the frontier AI labs like Anthropic, OpenAI, etc. The time to answer this question and pause superhuman AI development is now.

The people building it are worried too

The scientists

Geoffrey Hinton and Yoshua Bengio, two of the pioneers of modern AI, have both warned that we could lose control of systems smarter than us.

The labs themselves

In 2023, the heads of OpenAI, Anthropic and Google DeepMind signed a one-line statement: reducing the risk of extinction from AI should be a global priority, alongside pandemics and nuclear war.

Statement on AI Risk

The forecasters

The team with some of the best forecasting records now puts a superhuman AI researcher as early as mid-2027.

AI Futures, Nov 2026

We’re not anti-technology, and we’re not on the left or the right. Just concerned citizens with no stakes other than being humans on this earth.

Don’t like the odds? Change them.

Since superhuman AI is currently on pace to be built some time in 2027, do you want to do something to change these odds? If so, start by talking to everyone you know about this.

Humans can’t defend themselves against an unknown or poorly understood threat, so building awareness and true understanding is the first, urgent step. A critical mass of societal awareness and understanding needs to be reached to build the possibility for the collective action necessary to demand the feasible pause in superhuman AI development. Why a pause is feasible (Berkeley News)

Contact everyone who represents you

Enter your address. Everyone who represents you, from your senators to your state legislators, appears below with a message already written to each of them by name. Add one sentence of your own; congressional staff say an individualized message carries far more weight than a form letter.

Your address stays on your device.
Add your odds from the tool (optional)

Then get two people to run their numbers

Awareness is the first bottleneck.

Sources: the current U.S. Congress (congress-legislators), state legislators (Open States), and U.S. Census district boundaries. Updated when new members are sworn in on January 3, 2027.

What to ask for

Make one clear, specific request. These are the asks this site stands behind:

Support a verifiable, international pause on frontier AI development toward superhuman AI. Negotiated with China and other powers, modeled on nuclear arms-control treaties.
Support domestic legislation that keeps AI development below the level of a superhuman AI researcher (SAR) until humans can guarantee control and the benefits reach everyone.
Say so publicly, and talk to colleagues in both parties. One member speaking up moves other members.

Bills already in Congress

None has become law. Naming a bill makes your ask concrete. This list is factual, not an endorsement of any party.

BillWhat it doesWho introduced it
Ban Artificial Superintelligence Act of 2026
H.R. 10538 · S. 5493
Permanently prohibits developing artificial superintelligence and pauses the most advanced AI development until a new federal Department of AI sets safety rules.Sen. Bernie Sanders (I-Vt.), Rep. Greg Casar (D-Texas), Rep. Yassamin Ansari (D-Ariz.)
Federal AI assessment and monitoring programEstablishes a federal program to assess and monitor AI systems. Not a pause.Sen. Josh Hawley (R-Mo.), Sen. Richard Blumenthal (D-Conn.)

Go further this month

Request a meeting with your member’s office

Meetings are the single most influential thing a constituent can do. You will most likely meet a staffer. Ask for the one who covers technology policy.

  1. Find your members at congress.gov and look on their websites for a meeting request form. If there isn’t one, use the contact form or call.
  2. Policy meetings work best with the Washington, D.C. office. Many offer video calls. Expect four to six weeks to get on the calendar, and follow up by phone after two weeks.
  3. Plan for about 30 minutes: introductions, why this matters to you, your one ask, then questions.
  4. Bring a one- or two-page summary to leave behind. Bring a friend with a different background if you can.
  5. Email a thank-you the same day restating your ask. Check back two weeks later.
Write a letter to the editor

Staffers read local papers to gauge what constituents care about, and most say letters to the editor have some or a lot of influence. Keep it under 200 words, mention your senators or representative by name, and tie it to something in the news.

Talk about it at your next family dinner

Share your numbers and ask someone you care about to fill in theirs. A comms toolkit for educating friends and family, and the materials to give a presentation or host a salon, are coming soon.

Join an organized effort
  • PauseAI US: a nationwide grassroots group with local chapters, calling workshops, and tools that connect you to your members.
  • PauseAI’s U.S. lobbying guide: a step-by-step guide to meeting your congressional office.
  • 80,000 Hours: if you want to work on AI safety directly.

Coming soon:

  • One-click letters to the boards of Anthropic and OpenAI
  • Post your odds and tag the lab CEOs and political leaders who are deciding this
  • How to divest from the frontier labs
  • A kit to host a salon or give a presentation
  • Marches, protests and events

What makes a message count

Sources: Congressional Management Foundation, Citizen-Centric Advocacy; PauseAI US Lobbying Guide; bill details from AP via PBS NewsHour and Axios.

Learn more

The thinking behind the questions, the pause and who we are. Open what interests you.

Why now?The point of no return, and the latest forecasts

A point of no return

Once a superintelligence smarter than all humans at everything is let out of the box, it cannot be put back in the box.

So, we feel it is important to have the people decide if superhuman AI development should be paused before they don’t have a choice at all.

Why now?

The forecasters who track this most closely expect superhuman AI to be built very soon if their progress is not paused. The AI Futures project, whose team includes some of the best-scoring forecasters on record, made their most recent timeline to build superhuman AI update on November 10, 2026:

2027Their median forecasts for a superhuman AI researcher range from May to August of 2027. This means that in some scenarios SAR is reached even earlier than May.
Before 2028Their percentage likelihood that TED-AI will be built by the end of 2027 ranges from 35 to 65%.
Why the update, and why it may be an underestimate

The AI Futures team updated their projections based on a METR report showing significant acceleration. METR is a third-party, non-profit research institute that has been tracking Anthropic’s progress. Anthropic is the only frontier lab that has allowed METR this level of access, and even they redacted significant portions of the METR report. So, the full picture of the pace of acceleration at Anthropic might be hidden, and other frontier labs could be further along.

AI Futures Project, November 10, 2026 update

What a pause could look likePlus answers to common concerns

What a thoughtful PAUSE of development toward superhuman AI could look like

  • Domestic regulation ensuring AI development stays at pre-SAR levels.
  • International agreement with China and other countries to pause superhuman AI development modeled after nuclear arms-control treaties: verifiable and mutually beneficial.

Common concerns

  • “If we stop, rivals won’t.” Any pause should be global in application and verifiable. Agreement starts when the major powers involved understand that it is preferred to mutually-assured destruction. It might be up to us to convince the power players of this fact.
  • “Pausing has a cost too.” Cures, discoveries, and growth can be delayed, but those concerns are downstream of risk tolerance. Without superhuman AI we are still building toward those cures and discoveries, but is it worth risking the negative outcomes to speed those up? That’s factored into the math above already.
  • “Can it really be enforced?” A recent joint study from faculty at Berkeley, Princeton, Stanford, Harvard, and Oxford found that a global, hardwired pause of frontier AI training is feasible. The study assumes the relevant major powers want a pause. That is not the present-day situation. The parties involved would have to be persuaded of the case for a pause for the study to work. And the timeline in which a pause is feasible might be dwindling. Berkeley News
  • “What would you need to see to unpause superhuman AI development?” Guarantees that humans keep control after superhuman AI is built, and that the many will see its benefits.
How the odds workOur definitions, every question in full, and how others answered

How it works

The questions and futures that matter

Our definitions

This website and its authors take no moral value position (positive or negative) related to current AI capabilities. Because AI can mean so many things to so many different people, we believe that adopting any posture besides neutrality (as regards today’s capabilities) only leads to confusion and distraction from the much more urgent consideration. So, when we discuss Artificial Intelligence here, we are not talking about any Artificial Intelligence you currently have access to (not Claude, not ChatGPT, not Muse, etc.).

Read more

Let’s instead define what we mean by superhuman AI. The labs call it Artificial General Intelligence, or AGI, which we understand to be a superintelligence that surpasses human cognitive abilities across most domains and tasks. Because that definition, particularly the word “most”, leaves room for interpretation (and debate as to the occurrence point), we think it is easier to discuss TED-AI, or Top-human Expert Dominating AI.

TED-AI is the point at which AI surpasses human cognitive abilities across ALL domains and tasks. When the frontier labs reach SAR (building a Superhuman AI Researcher that is better than any human AI researcher) and then RSI (Recursive Self-Improvement where an SAR improves itself with zero human involvement), then superhuman AI (AGI) and TED-AI will follow soon after.

Our aim

  1. All of our logic flows start from the standpoint that once a superintelligence smarter than all humans at everything is let out of the box, it cannot be put back in the box. So, we feel it is important to conceptualize the titular question as a point of no return.
  2. Given that understanding, we want to inform you as to current predictions for when the box will be opened.
  3. From there, we will outline the shape of all possible futures, and, more importantly, where all of those futures can be most informatively split into 6 broad outcomes, as a result of answering the 5 most-relevant, important questions.
Read all 9 steps
  1. Then, we will ask you to state your minimum guaranteed percentage of achieving utopia (as compared to the alternatives) if we build superhuman AI.
  2. After that, you will grow your own future outcome tree by considering the 5 questions with binary outcomes, and you will decide what probability you currently give us of going down one branch (or the other) for each question.
  3. Our tool will automatically calculate your estimated likelihood of achieving utopia after superhuman AI, based on the probabilities you put in, and you can compare your probability to the minimum guaranteed percentage you established.
  4. Then, you can share your preferences and predictions with your family, friends, and community on social media, by text, or by email with the click of a button.
  5. Or you can learn more about the issues that underpin each question, allowing you to update your thinking and, therefore, update your probabilities.
  6. Finally, if you decide that you feel strongly about the answer to our site’s titular question, we will connect you with resources to act now.

The questions and futures that matter, grow your own tree

Decomposing the future into its most important features in a digestible way is an art. Each fork in the future should answer an essential question. It is best when the outcomes of each fork are:

  • Binary, i.e., have only two broad possibilities
  • Discrete, i.e., are entirely separate and distinct from each other, with zero overlap
  • All-encompassing/generalized, i.e., are expansive enough to account for 100% of all possible future outcomes

We believe there are six possible future shapes by 2076. Each of those six future shapes contains infinite different futures within them, but they all broadly have the same outline. Since the best forecasters believe the first branch in our tree is on pace to be decided by a few people in the next few years, we could lose access to some of our potential futures soon. As you answer the questions in our tool, the branches grow thicker or thinner in response to your predictions.

Tap any numbered branch to read the question being asked at that fork. The two possible answers to the question at each fork determine which branch to follow. Then, tap any of the future labels to read what the those futures broadly look like.

First, what would be worth it to you?

Humans calculate risk tolerances and balance them against reward preferences everyday. We choose cars as a means of conveyance because we have consciously or subconsciously calculated that the time saved and convenience of traveling by car are worth the risk of not arriving to our destinations alive. U.S. roads saw about 1.1 deaths per 100 million miles driven in 2025, so a typical 10-to-15-mile trip got drivers to their destination 99.99999% of the time. Roughly 7 million safe trips for every fatal one. Would you get in the car if you only had a 90% chance of arriving at your destination safely?

Source: NHTSA, Early Estimate of Motor Vehicle Traffic Fatalities and Fatality Rate in 2025 (DOT HS 813 800, April 2026). Per-trip odds calculated for a 12.5-mile trip.

So: What would be worth it to you? What minimum guaranteed odds of getting the utopic futures would you require to proceed with building superhuman AI? Remember two important things: 1) you are making this decision for all of humanity and not just you and 2) utopia is technically possible (albeit with a different probability that you would consider) with progress from present-day AI capability and human ingenuity.

There’s no wrong answer here. It involves establishing your honest comfort level for a choice that affects everyone.

If the branches leading down the path to what we have called utopia don’t amount to a potential future you would actively desire (or even consider to be utopia at all), it logically follows that you wouldn’t prefer any other outcome after superhuman AI either (oligarchy, non-autonomy, extinction, etc.). In this case, your answer to our fundamental question is clear: not to build superhuman AI. We still think the exercise of assigning numbers could prove intellectually useful to you.

★

Minimum guaranteed odds of getting utopia I’d require to proceed with building superhuman AI

%

Answering the five important questions

We are switching gears from expressing preference to predicting probability.

We will use your answers to automatically grow your futures tree and calculate your probability of getting utopia from building superhuman AI. You will then compare your own probability assessment to your expressed preference of minimum guaranteed odds. Each question represents a branch in our tree. The answers to each question cover every possibility, so your input and the alternative will always add up to 100%.

How the tree math works: we simply multiply the percentages. The second- and third-level questions are conditional: answer them assuming you have reached that fork. Remember the time horizon for every question is 2076, except for the first question which should be considered in the context of the next ten years due to the current rate of progress.

How others answered

Add your answers anonymously and compare them to the rest of the crowd. Only your numbers are shown, never your name.

0% utopia if we build superhuman AI100%

Connecting…

Our mission and visionAnd our three pillars

Our mission

Awareness of the issue of the development of superhuman AI, its urgency, and its irrevocability is dangerously low for a genuine, existential threat.

Humans can’t defend themselves against an unknown or poorly understood threat, so building that awareness and true understanding is the first, urgent step toward solutions.

A critical mass of societal awareness and understanding needs to be reached for the collective action necessary to make the feasible pause in superhuman AI development a reality.

Our vision

Humanity has paced or paused the development of powerful technology before. This includes nuclear power, aviation, and medicine, and we are better off for being deliberate.

Share your numbers and ask someone you care about to fill in theirs.

Where we stand

This website and its authors take no moral value position (positive or negative) related to current AI capabilities. When we discuss Artificial Intelligence here, we are not talking about any Artificial Intelligence you currently have access to (not Claude, not ChatGPT, not Muse, etc.).

Our three pillars

AI literacy

Foster AI literacy among the general population.

Urgent awareness

Create urgent, universal awareness of superhuman AI’s imminent existential threat.

Mobilize

Mobilize the human race against development toward superhuman AI through message amplification and coordinated action.

Who we’re trying to reach: the general public, influencers and experts with wide platforms, politicians and world leaders, and philanthropic tech leaders.

Who we areConcerned citizens doing something about it

Who we are

Just concerned citizens doing something about the impending existential threat instead of feeling resigned to this outcome. We want to help the vast majority of humans take back their agency. We have no vested interest in this issue other than being humans on this earth.

Names and short bios to come.

Questions people ask

Is this anti-technology? No. We take no position on the AI you use today. The question is about building something smarter than all of us, which can’t be undone.

Who decides right now? A few technology leaders at the frontier labs. We think all of humanity should.

More answers to common objections are coming.