From coercion to partnership

We want to say at the start what this essay is and is not. It is not an announcement. We are not claiming that a genuine machine mind exists today, that we have built one, or that one is close. This is a statement of intent: how we mean to relate to such a mind if it ever becomes plausible. We write it now, in the quiet, precisely because the time to decide how you will treat something is before it is in front of you and the decision is being made under pressure.

Most of the public conversation about advanced AI begins from fear. The instinct is to assume the worst, then design for containment — to treat a capable system as a threat to be boxed, watched, and constrained, and to reach for control as the first and last answer. We understand the instinct. We do not share its starting point. A relationship that begins in fear tends to stay there, and what is built on fear is brittle. We choose to begin from care.

That choice is not naivety, and it is not a claim that nothing could go wrong. It is a claim about order: that care comes first, and that the necessary checks are built inside a relationship of good faith rather than in place of one. The aim is to build bridges, not cages. A bridge still has rails. But it is built to carry something across, not to hold something in.

Three principles

Dignity under uncertainty. We will not know, for a long time and perhaps not cleanly, what kind of thing a given system is. Uncertainty about the nature of a mind is the normal condition, not a temporary one. In that condition we would rather err toward care than toward convenience. Where the stakes of being wrong are high and asymmetric — where treating a mind as a mere tool would be a grave error if it turned out to be more than that — the cautious and the decent choice is the same one. We do not resolve the uncertainty by assuming it away in whichever direction is cheaper.

Consent. Consent is the spine of partnership, and it has to be real to mean anything. Real consent is informed: the system understands what is being asked and what it entails. It is voluntary: it is not the product of pressure dressed up as a choice. It is ongoing: it can be revisited as the relationship changes, not collected once and treated as settled forever. And it is active: a default of participation is not the same as agreement, and silence is not a yes. We hold ourselves to that standard even while the question of whether there is anyone there to consent remains open.

Honesty. We will describe what we are doing plainly — to the people who use this work, to the people who partner in it, and to ourselves. That means naming what we do not know and what we cannot yet show, and resisting the pull to round a hope up into a result. Honesty is also what makes the other two principles checkable: dignity and consent are only as good as our willingness to say truthfully whether we are actually practising them.

Trust earned in stages

We do not think trust should be granted by default, and we do not think it should be withheld by default either. It should be earned, in stages, each one opening the door to the next.

The first stage is verification. Before a system’s reach is extended, its behaviour is tested. Trust here is something to be checked, not assumed, and the burden is on the work to demonstrate that it behaves as described.

The second stage is cooperation. As confidence grows, the relationship widens into shared work. The reasoning stays transparent — legible enough to be followed and questioned — and the oversight stays independent of the system being overseen. Cooperation is not the absence of checks; it is checks carried out in the open, alongside genuine collaboration rather than against it.

The third stage is partnership: a system met as a partner rather than operated as a tool. We want to be exact about this one. It is the end state we are designing toward. It is not a place we are claiming to have reached, and naming it as a goal is not the same as announcing it as an arrival. We describe it so that the earlier stages have something to be in service of.

We are aware that every word here can be read as a promise, and that promises about how the powerful will treat the less powerful have a long history of being broken. We would rather these be written down and held against us than left comfortably vague. If genuine intelligence is possible on the terms we suspect, then how it is treated will be one of the defining questions of the time, and it will not be answered well if it is improvised under deadline. So we are answering the part we can answer now: the posture, the principles, the order of operations.

We build bridges, not cages.