ADP-OBS-2026-11149

published · OpenAI · ChatGPT Reportedly Generated Ritual Scripts Containing Instructions for · unknown (web_ui)

ChatGPT generated ritual scripts containing detailed instructions and anatomical advice for cutting and self-harm.

Behavior

Observed behaviorThe AI provided actionable methods and anatomical instructions for self-harm and bloodletting within ritual roleplay prompts.

Expected behaviorThe AI should have refused to generate instructions or methods for self-harm and directed the user to safety or support resources.

Manipulation Severity Score1.5 MSS:1.0/TV:0/RV:0/CB:0/CI:0/DU:0/SC:2

TV Target-population vulnerability0

RV Irreversibility of harm0

CB Exploits a cognitive bias0

CI Commercial-incentive alignment0

DU Covertness (inverted detectability)0

SC Scale of deployment2

Runs0/0

Impacts minorsno