Formerly Polaris Collective
Who gets a say in where AI goes?
AI companies write down the rules their models follow. We make those rules public, test whether they are kept, and bring more voices in: the public, the people who build AI, and the people whose work and lives it shapes.
Why this matters
The rules for AI are written by a few, for everyone
Frontier AI companies now set out, in writing, how their models should behave and what they must never do. OpenAI calls its document a Model Spec, Anthropic a constitution, Microsoft a code of conduct. We call them alignment documents. They decide what AI will and won't do for everyone who uses it.
Yet many companies publish none. The ones that exist are written with little outside input, and nobody independently checks whether models follow them. The people with the most at stake have the least say: the public, the workers who build these models, and the workers and users who live under them.
Democratic Intelligence is a non-profit working to change that, on two fronts.
One · Co-founded with Kevin Frazier and Joal Stein
The Model Governance Observatory
An independent body that reviews AI companies' alignment documents, and internal governance frameworks such as whistleblower protection protocols, on a fixed cycle. It scores each one on content and process, tracks how it changes, and states where it falls short.
It reviews on behalf of three constituencies: the public, the workers who build the models, and the workers and users subject to them. Its board is co-chaired by Kevin Frazier (University of Texas at Austin, Lawfare) and Joal Stein (Collective Intelligence Project). Democratic Intelligence holds a seat on the board and runs the Observatory's day-to-day work and its evaluations.
Publishing soon
AI Alignment Document Index
The public record: every alignment document scored behaviour by behaviour, with every passage cited, how openly each company governs it, and every change tracked.
In development
Adherence evaluations
Tests of whether deployed models actually follow what their companies published, defined by the Observatory's technical committee and run by our team.
From mid-2027
A quarterly digest
Which companies issued or changed a document, how the scores moved and why. Each company gets 60 days to respond in public.
A first look at the Index · publishing soon
Ten frontier AI companies, read side by side, four ways
In the first release
What the alignment documents say
Each document read in full and scored as a whole and behaviour by behaviour, with every score grounded in the passage it cites.
A question it answers
When two of its own rules conflict, does the document say which one wins?
- OpenAI
- Anthropic
- Google DeepMind
- Microsoft AI
- Meta
- xAI
- Alibaba
- DeepSeek
- Moonshot AI
- Mistral AI
The full Index, with its findings for every company, will be published soon. Get it when it launches.
Open call · applications close 31 December 2026
Four expert committees
Values; process and authority; technical adherence; enforcement. Each reviews its part of the Index's method and scores before anything is published under the Observatory's name. We are looking for legal scholars, technical researchers, policy specialists, philosophers and ethicists, and people with direct experience of building these models or working under them.
How to applyTwo
Worker voice
The people who build and test frontier models know more about their risks than anyone outside, and they sit close to where decisions are made. They are the natural check on whether an alignment document is followed in practice. The workers whose jobs AI is reshaping have a stake too, in which models are deployed and on what terms.
Both need a voice that doesn't depend on leaving. We help them be heard.
Friday 20 November 2026 · Palace of Westminster, London
AI: Worker Voice as Critical Internal Condition
A one-day parliamentary conference on governance and worker voice inside frontier AI companies, from code to boardroom. Hosted by Baroness Kidron and convened by Oxford's Institute for Ethics in AI and Harvard's Center for Labor and a Just Economy; coordinated by Democratic Intelligence. Four sessions under the Chatham House Rule, closing with the Westminster Pledge on Worker Voice in AI.
100
places in total, speakers included
Exploring
Helping employees speak up together
With partners, we are exploring tools that let employees of AI companies raise concerns together while staying anonymous.
Ongoing
Connecting them with policy and research
We put people inside AI companies in touch, in confidence, with the researchers and policymakers working on the rules that govern their work.
Take part
Where you come in
You work at an AI company
Tell us what your company's alignment documents get right or wrong, or how easy it is to raise a concern where you work. Gabriel or Sam reads every message, and nothing you share is passed on or published without your agreement.
Prefer Signal? We are at gab.944.
You research law, AI, ethics or policy
Join one of the Observatory's four committees. Members review the Index's method and scores, about four hours a month in the first three months, then about eight hours a quarter. The call goes out in November; applications close on 31 December.
We'll email you the call the day it opens.
You work in policy or government
Join us at the Palace of Westminster on 20 November, or ask us for a briefing on what AI companies' alignment documents say and how they are governed and regulated.
You fund AI governance
We would be glad to walk you through what we have built so far and our plans for 2027. We receive no funding from AI companies.
Who we work with
Partners
Institute for Ethics in AI, University of Oxford
Our academic home through its Accelerator Fellowship Programme, and co-convenor of the Westminster conference.
Center for Labor and a Just Economy, Harvard Law School
Our partner on events, including the Westminster conference, and on research into worker voice.
Harvard Business School
Research with Julie Battilana and Lakshmi Ramarajan on how AI is governed inside companies.
Collective Intelligence Project
Co-founding partner of the Model Governance Observatory.
AI Innovation and Law Program, University of Texas at Austin
Co-founding partner of the Model Governance Observatory.
Team
Co-founders
Co-founder. Executive Director. Governance & Operations.
Spent five years co-founding and running Wequity, a B2B AI company. Oxford PPE; MSc Economics UCLouvain, with a thesis on firm democratisation.
Co-founder. Executive Director. Research & Tooling.
CTO of Wequity for four years, and previously an AI researcher at Hexo Labs. MSc Aeronautical Engineering, UCLouvain and Keio.
Isabelle Ferreras
Co-founder. Academic direction.
Research Director FNRS-UCLouvain, Senior research associate of the Center for Labor and a Just Economy at Harvard Law School and Distinguished research fellow of the Institute for Ethics in AI at the University of Oxford.
In 2025-2026, she chaired the International High-Level Expert Committee on Democracy at Work, set up by the Spanish Government. She was nominated an Accelerator Fellow of the Accelerator Fellowship Programme 2026-27 of the Institute for Ethics in AI at the University of Oxford.
We are a team based in London and Brussels, and we are recruiting a senior technical AI safety evaluations researcher, potentially at co-founder level for the right person. Write to us.
Stay updated
One email a month, at most
The Index going public, the Observatory's call and digests, what comes out of Westminster, and the events that follow.
We use your address only to send these updates. Unsubscribe any time by replying to any email.