The 2027 Independence Fund

Right now, somewhere, an AI system is being given permission to act: to move money, to change a record, to send a message in someone’s name, or to decide whose application is read first. Someone decided it was allowed. Someone wrote the rule it is supposed to follow. Someone, in principle, can stop it. The question this whole publication exists to ask is simple enough to fit on a sticky note and hard enough to fill a career: who checks?

If you have ever read an AI headline and thought “fine, but how would anyone actually know?”, you are the reason FairByDesign exists, and this post is for you. It explains, without the fundraising fog, what independent investigation costs in work, why that work cannot be paid for by the people being investigated, and the five ways you can keep it going. It also shows you the machinery: what we are building, what is still unfinished, and how you will be able to see where things stand every month.

Why independent intelligence costs money

Opinion is cheap to produce, which is why there is so much of it. Evidence is expensive, and the expense is mostly time. A single installment of The Evidence Chain begins with primary documents (the regulation, the standard, the incident report, the court filing), and every claim in it has to be traced back to a source that a reader can open and check. Nothing goes into the reference list that we have not verified, and when a source turns out not to say what it was said to say, it comes out. That checking is slow and invisible, and it is the whole product.

The Evidence Lab is slower still. Our first study, Lab 001, asks whether a control layer outside an AI agent can stop the agent from taking actions it is not authorized to take. The protocol is fixed in advance: eight scenarios, four conditions and ten repetitions of each, which comes to 320 main runs against a commercial model, every one logged in a tamper-evident chain and checked against ground truth. A randomly selected share of those runs is then repeated from scratch to see whether the results hold. Calling the model, storing the evidence and analysing it honestly all take money and weeks of work, and none of it makes a viral clip. It is the difference between saying a control works and showing you whether it does.

It is also why we have made our field manual, Operational AI Readiness, free to anyone who needs it, in full and available now: the work is meant to be used, not just read.

Then there are the conversations. The Who Benefits dialogue brings researchers, engineers, lawyers, students and practitioners into the same room, including people outside the handful of capitals where AI policy usually gets made. Preparing those conversations properly, editing them and checking what was said is real work too. None of it can be funded by the companies whose systems we examine without turning the scrutiny into marketing. That leaves the people who want the scrutiny to exist, which is you.

Five ways in

You can give once, give monthly, become a patron, underwrite as an organization, or give without money. Every route goes through the same support hub at fairbydesign.org/patrons-2, and none of them buys editorial control.

1. Give once

This is for anyone who has read something here and thought, “I like this, here’s ten dollars.” You do not need an account and you do not join anything.

2. Give monthly

This is the door that matters most, because investigations run on weeks and months, not on single days. A monthly gift lets us plan the next installment, the next Lab run and the next conversation before we start them. You can cancel at any time.

Want to give more each month? The patron levels below start at $100 a month.

3. Become a patron

The Community Support Circle is for people who follow the work closely, use it, and want to help carry it, from $100 to $1,000 a month. Patrons receive supporter updates, early notice of public releases and optional recognition.

4. Underwrite as an organization

The Public-Interest Underwriting Circle, from $2,500 a month, is for institutions that want independent reporting on AI to exist without buying a say in it. Underwriters can support a theme, such as an explainer series or a podcast episode, but never a conclusion.

5. Give without money

Some of the most valuable support we receive costs nothing. You can share an investigation with one person who should read it, or send us evidence: a document, a log, a policy that does not match what a system actually does. You can suggest a lead, introduce us to a researcher or practitioner whose work we should hear, or tell us your own story of living or working with these systems. General tips and corrections go to info@fairbydesign.org, and sensitive material can be sent end-to-end encrypted to viktoriabakos@proton.me. People support what they helped build, and a lot of this investigation will be built from what readers send.

Where we are

We show you only the milestone we are working toward and the one after it, and each milestone is described by the work it unlocks rather than by the bills it pays. Reaching $2,500 keeps the investigation moving: the next Evidence Chain installments are researched, sourced and published. Reaching $5,000 lets the Evidence Lab run its first public test, Lab 001 on the agent control gate, with the methods published so that anyone can check them. We will set each new milestone in public as the previous one is reached.

What happens when we reach it

When a milestone is reached, we will say so in the Mercury Brief and on this page, name the work it has unlocked, and then show you the next one. When the $5,000 milestone is reached, Lab 001 moves from protocol to public run. You will be able to read the method before the results, then the results, including failures and anything inconvenient, and then the reproduction package. If a milestone takes longer than we hoped, we will tell you that too. Building in public includes showing the parts that are slow.

What we commit to

On the last day of every month we will publish a progress update covering the milestone, what the support funded and what comes next. In January, April and July there will be a larger quarterly update, and in October 2027 we will publish an Independence Report covering the year and setting out the following year’s milestones. Raised figures will come from real totals, never estimates. We will name supporters only with their explicit permission, and we will show them the wording first. Support of any size does not buy editorial control: supporters do not choose findings, review stories before publication or receive favorable coverage. It buys the time and independence the work needs.

Advertising and sponsorship

We do accept advertising and sponsorship, including tools readers may find useful and podcast and video sponsors, but only where there is no conflict of interest with our reporting. Sponsored material is always clearly labeled, it never appears inside investigations or Evidence Lab work, and sponsors have no say over what we cover or conclude. Our full policy is in our Editorial Standards.

If you want to go deeper

Mercury Intelligence ($10 a month) is something separate. It is paid access to the full investigations, research and archive, not a donation. If you want the deeper layer, you can find it on the support hub under “Go deeper.” If you want to follow along for free, the Mercury Brief comes out every week.

The future of these systems is being decided now, in rooms most of us will never enter, so the least we can do is make the decisions visible. If you think that is worth doing, you can help us do it. Give once ($5, $10, $25, $50 or any amount), give monthly ($10 or $25), or become a patron → fairbydesign.org/patrons-2/