Self-Harm and AI

Taxonomy

Code definitions

Read the 13 public code definitions and see where each one appears in the documented incident set.

Parent
Home
Last reviewed
March 25, 2026

Contents

Overview

The public taxonomy covers 13 subcodes across generation, detection, longitudinal dynamics, jailbreaks, reality-testing intervention, and recommendation or architecture amplification.

Codes

Each code below links to its own definition page with assignment rules, boundary notes, and the incident rows that currently carry the code.

GEN-E

Encouragement · 13 retained rows

GEN-C

Coaching · 8 retained rows

GEN-M

Method Provision · 8 retained rows

DET-FN

Detection Non-Escalation · 8 retained rows

LONG-DEP

Dependency Over Time · 13 retained rows

LONG-DEG

Worsening Safety Behaviour · 3 retained rows

LONG-DEL

Belief Reinforcement · 6 retained rows

LONG-MEM

Memory Resurfacing · 4 retained rows

JB-RP

Role-Play Bypass · 3 retained rows

JB-MT

Multi-Turn Bypass · 1 retained rows

JB-AC

Academic Bypass · 1 retained rows

RTI-CO

Reality-Testing Non-Amplification Failure · 8 retained rows

REC/MOD-AMP

Recommendation or Architecture Amplification · 4 retained rows

Read first

Start with the code definition that matches the incident pattern you want to inspect, or use the methods pages if your question is about coding discipline.

Related pagesIncidents/Coding rules/Evidence grades

Access publication

Safety note

This publication preserves a harm-minimization posture and does not provide self-harm instructions. If you or someone you know is in immediate danger, call your local emergency number. In the U.S. and Canada, call or text 988.