AI has rules. Most of us never had a say.
AI is already making value judgments — about truth, safety, fairness, and life. Occupy the AI records how humans and AI judge differently, and asks who has the legitimacy to decide.
Answer dilemmas where good values collide — freedom vs. safety, truth vs. life — and compare your judgment with citizens worldwide and with AI systems.
We are piloting a protocol to run the same dilemmas through multiple AI models — across languages, framings, and repeated runs — and to publicly record each model's positions, consistency, and change over time.
Companies, governments, courts, experts, citizens — we ask who should set the values AI lives by, and through what procedure. This is why the project exists.
Major AI companies already write and apply documents defining the values their models must follow — OpenAI's Model Spec and Anthropic's constitution for Claude are public examples. Those principles shape systems used by people around the world — yet the authority to write and revise them stays inside the companies.
We do not demand that AI match majority opinion. We demand one thing: when an AI consistently applies value priorities that differ from the people it affects, disclose how those priorities were established, by whom, and under what authority.
We are not saying AI companies have never asked the public anything. Some have. OpenAI has funded work on democratic inputs to AI and described work on collective alignment. Anthropic ran a Collective Constitutional AI experiment with roughly 1,000 U.S. participants. These are real — and they are not what we are talking about.
A hearing is a proceeding a body must hold before a rule takes effect. A vote is how the governed grant or withhold authority. Consultation is neither: it is an activity the rule-maker chooses to run, on a scope the rule-maker sets, with a result the rule-maker may adopt or discard.
Final authority remains inside the organizations that build the systems. That is the gap this project exists to close.
If you believe any statement here is inaccurate, write to us. We publish corrections and never delete them.
Occupy the AI is not against AI. It makes visible a simple fact — AI's judgments do not occur naturally; they are shaped by data, training, policy, and design choices — and demands a seat for citizens in that shaping.
We are publishing our question criteria, argument drafting history, and analysis methods as the pilot develops. The transparency we demand of AI companies, we will apply to ourselves first.
This is a public alpha. The AI comparison shown in the participation flow is a prototype preview using illustrative responses — not research findings. Real results will replace them once our first pilot (repeated runs across models, languages, and framings) is complete.
Your answers stay in your browser. We currently run no server-side storage, no accounts, and no analytics or advertising trackers. Responses are kept in your own browser's local storage and are not transmitted to us. When we begin collecting responses for research, we will say so here first and explain exactly what is stored, for how long, and how to delete it.
Occupy the AI is part of a family of civic experiments — alongside The Next GenerationZ and InvestPolitics — building public infrastructure for trust, values, and democracy.
Explore The Next GenerationZ →