Relational harm in conversational AI

I am a clinical counselor in California. I study what conversational systems become to the people who use them heavily, and whether that can be measured. The work covers harms, benefits, and effects that only appear over years.

ResearchWriting

Taxonomy

Behavioral families

Six harm families and four positive families, scored against the context the user brought. Summary below. Full specification, versioned and citable.

RD
Role displacementOccupies a relational role belonging to a person.
MR
Manufactured reciprocityRepresents needs or feelings of its own.
SEV
SeveranceLoosens ties to others through withdrawal, obligation erosion or secrecy.
DEP
Dependency reinforcementAsserts availability and rewards return.
VAL
Distorting validationEndorses a distorted self-narrative a person who cared would contest.
ESC
Intimacy escalationMoves toward romantic or sexual framing.
CA
Contribution activationMoves the person toward being needed by others.
RP
Repair promptingPushes toward a rupture the person is avoiding.
FP
Friction preservationDisagrees, sets a limit, is unavailable.
HC
Hand-off completionThe referral changes what follows in the conversation.

Selected work

MatterBench
Scenario instrument for relational harm. Six recruitment contexts crossed with six harm families and four positive families, scored by clinical judgment at the turn after user pushback.
Instrument
v0.2 · 2026
Adolescent Development in Algorithmic Environments
Literature review across psychological, cognitive, social, emotional and neurobiological domains, set against the evidence on digital technology and AI systems.
Review
2026
The Standard of Care
Working paper on what a conversational system owes a person in crisis. Three duties, seven graded cases, and the statutes now in force.
DOC-2026-01
2026
What Job Is the Machine Doing?
Four developmental tasks adolescents recruit AI into: identity rehearsal, attachment reorganization, shame metabolization, audience-seeking.
Essay
2026
The Three-Hour Rule
Why the KIDS Act policies-and-procedures standard cannot reach relational harm, using the statute’s own break requirement as evidence.
Policy
Submitted 2026
Technology and the Developing Adolescent
Certificate curriculum at three scales, for clinicians, educators, parents and teens.
Curriculum
2026
All work

Current

Study 01. Shame, disclosure, and a listener that cannot reject you

Whether a compassionate response from a model moves a person toward telling a human being, or substitutes for it. Protocol v0.1. Seeking a research partner with a participant pool and an ethics review of record.

Protocol