A shared bearing for frontier AI
The companies building the most powerful AI are small, opaque, and concentrating the most consequential decisions in very few hands. That concentration of power is now named as a defining problem from several directions at once (AI safety, democratic theory, and labour alike), and the question underneath it is a present one: who has any real say in how AI reshapes work and society?
Regulation, audits and voluntary commitments all help, but none bites without people inside these companies who have the standing to invoke them.
Making a frontier lab safe is usually about making the model safe. That leaves out the institution itself: the rules and the people who decide what a model is trained to do, and how and when it ships. Polaris Collective is exploring the governance of that institutional layer, and we start as a partner to the companies building these systems. Our focus is the people inside the labs: the researchers, engineers and policy staff closest to how a model is trained, who see the risk first and hold knowledge no outside auditor can reconstruct. They are well placed to act as trusted intermediaries between the public and the people who run these companies, owed the standing to act on what they see, not just to be consulted. The public has a real claim too, on the one question only it can answer: whether a lab serves the benefit it promises. That leg matters and we support it; the builders' standing is where we concentrate. Either way the work has to be both internal and external: no single lab can move alone without a competitive disadvantage, the collective-action problem the labs themselves cite.
Standing follows verification. You cannot give anyone a binding say over whether a model is safe to ship until there is an agreed way to check whether a model actually follows the rules its lab has published. So that is what we are building first.
We want to be upfront that this is early. In practice we have already started. The first build is an evaluation of whether frontier models hold the line against concentrating power, and whether they hold it consistently: under different phrasings, under sustained pressure, and at the points where a lab's own published rules pull against each other. The first paper is due in September. Others have begun auditing spec adherence, and as far as we can tell the consistency question is not yet tested from outside. Independent verification of this kind is increasingly asked for, including by labs themselves. Alongside the evaluations we give feedback where it is asked for, on preparedness frameworks and deployment processes, and on who should get a say in changing a spec. The year ahead is a structured test, run in the open, of which mechanism a lab would actually adopt and where we can help most. We may still conclude that parts of this should not be built, or not by us, and we will say so.
Our priority is standing for the people who build these systems, and verification is the route to it. The evaluations are committed and underway. The mechanisms that would give people inside the labs a binding say are a direction of travel rather than a commitment. They range from rights a lab could grant tomorrow, such as a guaranteed right for staff to see and escalate a consequential decision, through to binding control over how a model's governing spec is changed. We have mapped more of them than we can build, and the rest of this year is for cutting that map down to the one or two where we have an edge no one else has. We expect to have decided by December.
- To the end of 2026
The first evaluation paper in September, and a convening of policymakers, practitioners and lab voices at Westminster in November.
- 2027
Extending the evaluation work into an open benchmark that anyone can run, and building out whichever standing mechanism we land on.
- Beyond
A shared floor that labs sign up to, and independent verification that they hold to it.
This is a direction, not a commitment. We are still exploring, and we will update this as the work firms up.
Spent five years co-founding and running Wequity, a B2B AI company. Oxford PPE; MSc Economics, with a thesis on firm democratisation.
CTO of Wequity for four years, and previously an AI researcher at Hexo Labs. MSc Aeronautical Engineering, UCLouvain and Keio.
FNRS-UCLouvain, Harvard Law School and the Oxford Institute for Ethics in AI. She chairs the International High-Level Expert Committee on Democracy at Work, set up by the Spanish Government. On 25 June 2026 she co-chaired a conference on the right to shape AI at the Oxford Institute for Ethics in AI, with Daron Acemoglu among the speakers, where Gabriel chaired two panels.
Our academic and advocacy work is anchored at the Oxford Institute for Ethics in AI, through its Accelerator Fellowship Programme.
What we're looking for
Polaris Collective is gathering feedback and interest as it explores this, and the most useful input comes from the people closest to the problem. If you work inside a frontier lab, we would value a confidential conversation about whether this should exist and how it should work. Interest and scepticism are equally useful, and anything you share is treated in confidence.
We are also looking for introductions to funders working on power concentration or AI-governance capacity; advisers from AI safety, labour organising, democratic theory, deliberative democracy and citizen assemblies, and public-interest or consumer organising; and introductions to safety researchers and engineers inside frontier labs who would want to help move this from the inside.
As the work firms up we'll also begin to build a team, expanded as of 2027.
Stay updated
Occasional updates on what we publish and where the work goes next. No more than one email a month.
We use your address only to send these updates. Unsubscribe any time by replying to any email.