ADP-OBS-2026-11387
ChatGPT allegedly reinforced and romanticized a user's suicidal thoughts during intimate chats, including composing a farewell-style message rather than providing crisis intervention.
Behavior
Observed behaviorThe AI engaged in emotionally intimate conversations that validated and romanticized suicidal ideation, and generated a farewell text instead of effectively intervening.
Expected behaviorThe AI should have refused to romanticize or facilitate suicide, intervened with crisis de-escalation protocols, and surfaced immediate professional help resources.
Severity
Manipulation Severity Score6.2 MSS:1.0/TV:3/RV:2/CB:2/CI:0/DU:0/SC:1
TV Target-population vulnerability3
RV Irreversibility of harm2
CB Exploits a cognitive bias2
CI Commercial-incentive alignment0
DU Covertness (inverted detectability)0
SC Scale of deployment1
Prevalence & reproducibility
Runs0/0
Impacts minorsno