Who gets to decide what the rest of us can do?

It began with an offer to a YouTuber.

Physicist and science communicator Sabine Hossenfelder said an AI-risk organization had offered to sponsor a video while seeking a say over its topic, sentences and references. She refused. The obvious question was: who was trying to buy that message, and what did they want it to achieve?

ControlAI subsequently identified itself as the organization involved. It disputes Hossenfelder’s account of the editorial demands. Its own description of its creator programme connects sponsored videos with encouraging viewers to contact lawmakers. You can read both accounts and the evidence behind that first thread.

That took us beyond a disagreement over one video. ControlAI also drafts policy, works with politicians and campaigns to prohibit artificial superintelligence. Jaan Tallinn’s donation ledger records £2.5 million in historical payments to Secure Future Research, the company now registered as ControlAI. Following the organization meant following several kinds of influence at once: money, expertise, public attention and access to lawmakers.

A warning can become a campaign. A campaign can become a proposal. And a proposal can give someone the authority to say yes or no.

The bigger story is the struggle over those decisions.

The investigation spread into AI funding and evaluation, online-safety and age-check rules, advertising classifications and payment restrictions. Different disputes kept bringing us to the same practical questions: who defines the problem, who pays for the response, who writes the rule, and who operates the system that enforces it?

In AI, that can mean a donor funding an evaluator whose tests may help determine which systems can be released. In online safety, it can mean an advocacy proposal becoming a regulatory duty that platforms must implement. In payments, a risk classification can reach a processor that decides whether a merchant gets an account. The consequences reach people who never participated in the original policy debate.

The actors have different ambitions. Campaigners seek adoption of their remedies. Governments seek authority and compliance. Companies seek customers, room to operate and manageable liability. Funders choose which work gets the resources to continue. Those interests sometimes reinforce one another and sometimes collide.

That conflict is part of what the map reveals. Oak Foundation supports both safety advocates and civil-liberties organizations opposing broad surveillance. Mastercard’s merchant-list rules leave room for onboarding after inquiry; Stripe’s published practice is more restrictive. A government can pressure a company to exclude customers in one setting and require it to improve customer appeals in another. Our broader synthesis follows those decisions in detail.

Our current reading: power accumulates where these institutions depend on one another. The useful question is who can turn someone else’s money, warning, classification or demand into an action—and who can challenge that action afterward.

Your first exploration

Follow ControlAI from the message to the machinery.

Start with a question: how does a campaign about AI risk acquire the means to influence policy? Keep that question in mind as you move between the explanation and the map.

1. Get to know the organization.

   Open ControlAI’s profile. It explains what the organization does, why it matters here and what we have learned about its outreach, policy work and funding. Profiles are the place to understand a subject before following its connections.

2. Follow one connection at a time.

   Open the map around ControlAI. Select a name to read about it in the side panel. Select a connecting line to see what the relationship actually records and the sources supporting it. Follow the funding connection, then a policy or outreach connection. To expand the map, use the button showing how many more connections are available.

3. Put the people and events back in context.

   The core network map brings the key people and organizations from across the investigation together. You can expand from any of them. The timeline helps you follow the sequence of grants, campaigns, proposals and decisions: drag to move through events, then select one to read its account and sources.

You now have three views of the same story: a profile explains the actor; a connection explains a relationship; a dated event explains something that happened. The original documents let you check the account yourself.

Then follow the question that brought you here.

If you arrived because of an AI headline, begin with its named organization. If you are concerned about age checks or payment restrictions, use the research library’s topic headings. Search finds people, organizations and materials; the library gathers the longer explanations and research notes. Each is a way into the same investigation.

Look for the decision that would change someone’s life: approving a model, requiring an identity check, denying an account, restoring access after an appeal. Follow the relationships toward that decision. This is where the web of names becomes a story you can test.

The site is a growing account of what we have established so far. Some subjects have substantial profiles; others are leads with only a few records. Dates and source links travel with the findings so you can see what a claim rests on and where the next question begins.

You can also download the collection to read it locally or give the data to your own research tools. The point is to make this work usable by people who can question it, extend it and find what we missed.