A Say in AIDiscussČesky
Six AI companies committed to four safety controls. What would you require of them?
Read one sentence at a time. Agree, disagree, or pass. When you have voted on a few, write one of your own. We do not ask for an account or a name.
How it works
- You see one statement at a time and press agree, disagree or pass. Nobody can reply to anybody, so there is nothing to argue in and nobody to be shouted down by.
- We will switch on Polis’s opinion map. After a few votes it shows where you sit among everyone who took part, and which statements very different groups agree on. Those are the statements where people who usually disagree agree.
- We publish the results with how many people took part and what we cannot tell about them. We never present a count of clicks as the view of a country or of humanity.
The conversation
The first statements
Written by us, with help from Claude, an AI model made by Anthropic, to start the conversation. They deliberately include views we do not hold. The statements on slowing down and on AI that improves itself echo positions argued in a 2026 essay by Anthropic’s chief executive, Dario Amodei. Anyone can add their own once it is open. (34)
The accord
1The White House accord is a good first step.
2The accord is enough for now.
3Safety rules for the most powerful AI should be set by law.
4A company should not be allowed to choose its own auditor.
5The results of AI safety audits should be published.
6If a safety control at an AI company fails, the public should be told.
7Strict rules in one country would move AI development to countries with weaker rules.
Who decides
8People outside the United States should have a formal say in how the most advanced AI is built.
9Elected governments, not companies, should decide what the most powerful AI is allowed to do.
10Ordinary people chosen by lot should help decide the rules for AI.
11Rules for AI should be made by experts and elected officials, not by panels of ordinary people.
12A company’s board should include members chosen by the public.
13Decisions about the most advanced AI should be left to people with expert knowledge.
14The most advanced AI needs a treaty between countries.
Pace and limits
15Companies should slow down the most powerful AI until we understand how to make it safe.
16Companies are best placed to decide how to make their AI safe.
17Governments should wait to set rules for AI until the risks are better understood.
18If we slow down, less careful companies or countries will win.
19AI that improves itself without human review should be banned.
20A company should need permission before it trains a much more capable AI.
Your own AI
21My AI should always tell me what it cannot help me with, even if it cannot say why.
22My AI should never lie to me.
23It is fine for my AI to soften the truth to be kind.
24People should be able to choose the rules their own AI follows, such as how cautious or how blunt it is.
25AI companies should not train on my conversations without my clear agreement.
When an AI acts for you
26An AI that acts for me should ask first before it spends my money.
27An AI that acts for me should ask first before it shares my data.
28I should be able to withdraw an AI’s permission at any time.
29When I withdraw an AI’s permission, everything it was doing should stop.
30It is fine for an AI to act for me without asking each time, if I can see afterwards what it did.
31A group that uses an AI service should be able to set its conditions, not only the company that runs it.
32People affected by an AI they do not use should have a say in the rules it follows.
How we treat AI
33How people treat an AI system is not a moral question.
34AI companies should limit cruel treatment of their AI.