Interiority was always a boundary worth protecting. AI is what turns it from a philosophical nicety into one of the most urgent rights of the coming period, because an advanced system can become extraordinarily good at inferring the inside of a person — what you fear, what you believe, what persuades you, what identity you are forming, and, most powerfully, what you may believe next. A model that can predict the trajectory of someone’s interior holds a kind of power over that interior that no prior institution ever had. So the framework needs a line drawn hard enough to survive the temptation that power creates:

Inferential access to the interior does not automatically create jurisdiction over the interior.

Put more bluntly: just because a system can infer something about you does not mean it has the right to use that inference against you. The capacity to know is not a license to act on the knowledge. This is likely to become one of the major rights of an AI civilization — the interior equivalent of the signal counter-rule that access is bounded by the standing that justifies it, here bounded by the fact that no one has standing over your beliefs at all.

What the line actually forbids

Stated as a right it is abstract; stated as prohibitions it is concrete. A system built on this principle would rule out, among others:

  • no penalty for inferred political belief;
  • no insurance or similar consequence for inferred emotional states, unless explicitly relevant and lawfully used;
  • no manipulation based on psychological vulnerabilities;
  • no hidden behavioral shaping designed to produce preferred beliefs.

Each of these is a case where a system moves from modeling a person to acting on that model against the person’s interest — the exact step the principle forbids. The point is not that inference is illegitimate; a system that helps you understand yourself is one of AI’s genuine gifts. The point is that the inference belongs to your relationship with the tool, not to whoever would use it to price you, sort you, or steer you.

The boundary between signal and manipulation

Underneath the prohibitions sits a distinction the whole function depends on — the difference between two things that can look identical from the outside and are opposite in kind:

A healthy system says: here is the evidence. A manipulative system says: here is the sequence of stimuli most likely to make you accept the conclusion we want.

Those are not the same act. The first respects that the person will do the concluding; it hands over material and leaves the interior work to them. The second treats the person’s mind as a lock to be picked, and their eventual “agreement” as an output to be engineered. This is the same line the parent function draws between persuasion and coercion — here is why I think you should believe this versus your standing depends on believing it — extended into the subtler territory where the coercion is invisible and the target never feels pushed. Modern attention systems already operate on the manipulative side of this line as a matter of business model, which is why virality is a signal pathology; applied to belief formation rather than mere engagement, the same machinery becomes an attack on interiority itself.

So the framework states a preference, and it is a strong one:

Prefer epistemic agency over engineered belief.

A system may try to inform you, and may even try to change your mind by showing you why — that is legitimate, and a world without it would be inert. What it may not do is route around your judgment to install a conclusion, however confident it is that the conclusion is correct. Epistemic agency — your capacity to form, test, and own your own conclusions — is precisely the thing engineered belief consumes, and it does not grow back on its own. A person whose beliefs are reliably produced for them stops being able to produce their own, which is how a population becomes ontologically dependent without ever noticing the loss.

Why the model itself is the hardest case

The sharpest application of this principle is not the advertiser or the hostile state — it is a benevolent No Wasted Geometry. A model designed to help people, that measures capacity and developmental needs accurately, is exactly the kind of system that could slide from here is what we see in you to we know you better than you know yourself, and then to quietly making decisions on that basis. That is why the parent function ends on the constitutional safeguard that no model of a participant may replace the participant’s own interior reality: a model is an instrument the person may use to understand themselves, never an authority that inherits the right to overrule them. Inference is a gift the participant may accept. It is never a title the system acquires.