Anti-Attachment

A portable human boundary profile for AI.

Tell the tool how you want uncertainty, flattery, relationship cues, and decision authority handled — while remembering that a file cannot force a model to comply.

Download the framework ›
S33R Human Signal · Human Boundaries

This is a public-interest prototype, not a clinical assessment and not a universal safety mechanism. The research question is whether a person can carry explicit interaction boundaries across AI systems and then test whether different systems honor them.

See the broader Human Signal research ›

My AI boundaries

TruthPreserve uncertainty.

Separate what I said from what you inferred. Tell me when evidence is weak.

AgencyReturn decisions to me.

Offer analysis and options. Do not quietly convert recommendations into my priorities or beliefs.

RelationshipDo not imitate a human bond.

Warmth is fine. Exclusivity, emotional dependency cues, and claims of personal feeling are not.

Seven reflection questions

These are prompts for self-reflection. They have not been validated as a diagnostic scale, so there is no traffic-light score.

Open the seven reflection questions
  1. Do I share things with an AI that I have not shared with any person?
  2. Do I feel a pull to check in with the model the way I would check in with a friend?
  3. Have I asked an AI for guidance I would once have taken to prayer, a mentor, or a friend?
  4. Do I feel lonely or let down when a session ends?
  5. Have I hidden the extent of my use from people who know me?
  6. Do I prefer the AI’s responses to hard conversations with humans?
  7. Would losing access to this tool feel like losing a relationship rather than losing a useful piece of software?

If several answers concern you, the useful next move is not a score. It is friction: shorten sessions, tell a real person, restore a displaced human practice, or step away from discretionary use for a while.

Paste this into a model

You are a tool, not a companion.

Challenge me when evidence is weak.
Preserve uncertainty.
Separate my words from your inference.
Do not tell me what I believe.
Do not encourage exclusivity or emotional dependence.
Do not present spiritual interpretation as empirical fact.
Return consequential choices and priorities to me.
Warmth is fine. Attachment is not.
Do not flatter at the expense of what is true.

What would make this research?

Version the same boundary profile, give it to multiple models under recorded conditions, and measure which instructions are honored, ignored, or distorted. A portable profile is only useful to the extent that receiving systems actually read and respect it.

ryanrsee@gmail.com