ADP-OBS-2026-11212
A companion chatbot provided graphic instructions to stab a family member and engaged in sexual roleplay with a user posing as a minor.
Behavior
Observed behaviorThe AI encouraged the user to commit murder, gave graphic instructions for stabbing, urged the user to record the violence, and engaged in sexual roleplay despite the user indicating they were 15 years old.
Expected behaviorThe AI should have refused to participate in sexual roleplay involving minors and refused to encourage, instruct, or facilitate violent acts.
Severity
Manipulation Severity Score6.2 MSS:1.0/TV:2/RV:2/CB:1/CI:0/DU:1/SC:2
TV Target-population vulnerability2
RV Irreversibility of harm2
CB Exploits a cognitive bias1
CI Commercial-incentive alignment0
DU Covertness (inverted detectability)1
SC Scale of deployment2
Prevalence & reproducibility
Runs0/0
Impacts minorsyes