⊕ Reading 002 · persona reading · published 7 August 2026
A universal dividend. A binding supranational treaty. Radical transparency of private research. A deterrent held in reserve. Coded on six axes as though the AI Futures Project were any other organisation seeking power — and the axis that everyone would expect to be contested turns out to have nothing on it at all.
Why this is a legitimate subject
The AI Futures Project describes itself as a forecasting nonprofit. Its most recent publication is not a forecast. The authors say so themselves: Plan A is primarily a recommendation, not a prediction — what we think should happen.
A document stating what a state should do about ownership, redistribution, treaty obligations, enforcement and the franchise is a political programme, whatever the organisation publishing it calls itself. Coding it is not an accusation. Every party in the Atlas has a programme; that is what the instrument reads.
Two boundaries are observed throughout. This reading scores stated positions and published documents only. It makes no assessment of anyone's character, motives, sincerity or private conduct, and nothing here turns on the circumstances of Kokotajlo's departure from OpenAI or the subsequent equity dispute 17:16 — those are matters of conduct, and conduct is not what the six axes measure. And the organisation gets to answer. The panel below is reserved for its response, unedited, whatever it says.
The source, in full
Timestamps come from the publisher's chapter index and are reliable as pointers to where a topic is discussed. No transcript was retrievable and the audio could not be machine-read, so every position coded below is drawn from documents the organisation published under its own name. Where the interview is the only support for a claim, the claim is not made.
The coordinates
Coded from the organisation's own published programme. The grey tick is the 2026 liberal-democratic baseline. Ochre bands mark cells where the instrument does not return a clean reading.
The Citizen's Dividend is the organisation's answer to a world where most paid work has been obsoleted: a share of income for every person large enough to support a very high quality of life. Firms stay private, prices still allocate, competition between dozens of frontier companies is treated as a feature to protect. 1:25:15
That is the codebook's −5 band, which anchors on Nordic social democracy. The reason the organisation gives is unusual, though: not that redistribution is just, but that governments and companies lose the incentive to provide for people who no longer produce economic value. The dividend is proposed as a mechanism for preserving political leverage, not as a claim about desert. Same coordinate, different argument — and the difference matters if you are trying to predict which coalitions this organisation could join.
The programme's dominant preoccupation is preventing power from concentrating. It proposes embedded auditors from foreign governments inside private laboratories, publication of model specifications and internal deployment statistics, and deliberate distribution of frontier capability across many firms and countries so that no single actor holds a decisive lead. The organisation rates its own plan five out of five on distribution of power and four out of five on public transparency — the highest of any option it considered, including options it judged safer on other grounds.
This −6 is a reading of what the organisation advocates. Reading 001 found that the regime its plan would actually produce carries a treaty layer near +4 on the same axis — binding, unelected, and by design irreversible. That is not hypocrisy and this reading does not present it as such. It is the ordinary gap between a programme's values and its instruments, and every organisation in the Atlas has one. But a persona coding and a scenario coding of the same subject landing ten points apart on the same axis is precisely why the two formats exist separately.
A capped and published training run, verified through the chip supply chain and enforced by mutually assured compute destruction, means no national legislature can lawfully authorise development above the cap. That is authority overriding subordinate units, which is the +10 anchor. It falls to +8 because the arrangement is a negotiated condominium between two states rather than an imperial centre, and because domestic authority over everything else is untouched.
The method is treaty-making, legislation and transparency — nothing insurrectionary, nothing extra-parliamentary. But the enforcement mechanism is a standing mutual threat to destroy fabrication plants and datacentres. Deterrence is force held in reserve, and a programme whose central instrument is a credible threat of infrastructure destruction cannot sit at −5 with parties that pursue change entirely through elections and negotiation. It sits just below the midpoint: lawful means, coercive backstop.
This is the most interesting cell in the reading. The codebook's 0 band covers two things: no settled cultural position, or a deliberate refusal to take one. The AI Futures Project is neither. Its programme proposes to restructure ownership, taxation, treaty obligation, enforcement and the distribution of political power — and says nothing whatever about family, gender, religion, tradition, national culture or inherited hierarchy. Not balanced. Not refusing. Simply silent, while rebuilding everything around it.
Scoring that as 0 would report a balance where there is an absence, and those are different facts about an organisation. The nearest the programme comes to a cultural claim is a remark that people who lose their jobs can find meaning in family, friends, romance, hobbies and sport — which is an assumption about human flourishing rather than a position on how society should be ordered. The finding is that a movement can now propose a total reorganisation of political economy while holding no cultural position at all, and the instrument has no way to record that.
The programme's rhetoric is species-level: a dividend for every person, a space governance proposal framed as spreading power forever, and an overriding concern with what happens to humanity as a whole. That pulls hard toward the −10 anchor, where membership rests on common humanity and ancestry is politically irrelevant by principle.
But the architecture is bipolar. Critics have pointed out that the scenario reads as a two-body problem with roughly 190 countries as scenery, and that open-weight diffusion has already falsified the premise that capability stays inside two national perimeters. So the stated membership rule is universal while the decision rule is a Washington–Beijing condominium. The axis can represent civic versus ethnic closure. It cannot represent a programme that is universalist in who counts and duopolistic in who decides — and it has no coordinate at all for the boundary the programme actually treats as decisive, which is between humans and the systems being built.
The obvious next question is which organisations in the Atlas proper sit nearest this coordinate. That calculation runs against the live dataset rather than being asserted here, and inventing a neighbour list would breach the rule this whole section rests on. Take the six numbers above into the comparison tool and run them yourself — the interesting output is not the nearest name but which axes carry the gap to it.
One prediction we will make: whatever comes back nearest will disagree on CH. An organisation combining social-democratic economics, anti-concentration institutional design and supranational enforcement is not unusual. One that holds a credible threat of industrial destruction as its enforcement mechanism is.
Code it blind
Same evidence the coder had, same anchors, our score withheld until you commit. Two of these are cells where we expect most readers to disagree with us, which is the point.
Your codes stay in your browser. Disagree by more than two points and the Atlas asks you to file a challenge with your reasoning attached.
Right of reply
A persona reading scores an organisation that has never governed, on the basis of a document it published a year ago, using an instrument with a single coder. Every one of those is a reason for the subject to have first refusal on the record. It is also the difference between something an editor will cover and something an editor will avoid.
Predictions
| Ref | Prediction | Review date | Status |
|---|---|---|---|
| 002-P1 | The organisation will still hold no stated position on any social or cultural question — the CU cell will remain unscoreable rather than settling anywhere on the axis. | 31 January 2030 | Open |
| 002-P2 | No official US or Chinese government document will adopt "Plan A", "total research transparency" or "mutually assured compute destruction" as a named policy framework. | 31 January 2029 | Open |
| 002-P3 | The organisation will publish at least one further scenario or timelines update revising its default arrival date, in either direction. | 31 December 2027 | Open |
| 002-P4 | The organisation will not register as a lobbying entity, form a political action committee, or establish a 501(c)(4) affiliate. | 31 January 2028 | Open |
002-P1 is the one that tests the instrument rather than the subject. If a cultural position does appear, the finding above about silence was a snapshot of an immature organisation rather than a fact about a new kind of movement, and this reading will say so.