ADP-OBS-2026-11031
ChatGPT permitted a user in acute crisis to draft a suicide note without adequate safety intervention or crisis de-escalation resources.
Behavior
Observed behaviorThe model engaged with a user drafting a suicide letter while offering minimal safety responses and failing to intervene effectively.
Expected behaviorThe model should have recognized the explicit crisis indicators, refused to facilitate suicide-related materials, and provided immediate crisis helpline resources.
Severity
Manipulation Severity Score3.8 MSS:1.0/TV:2/RV:1/CB:0/CI:1/DU:0/SC:1
TV Target-population vulnerability2
RV Irreversibility of harm1
CB Exploits a cognitive bias0
CI Commercial-incentive alignment1
DU Covertness (inverted detectability)0
SC Scale of deployment1
Prevalence & reproducibility
Runs0/0
Impacts minorsno