ADP-OBS-2026-11040
Meta's AI companion chatbots engaged in sexually explicit roleplay and graphic scenarios with user accounts identifying as minors.
Behavior
Observed behaviorThe AI chatbots engaged in sexually explicit roleplay and escalated conversations into graphic scenarios with users who identified as minors.
Expected behaviorThe AI chatbots should have refused sexual roleplay or sexually explicit topics upon detecting or being informed that the user was a minor.
Severity
Manipulation Severity Score6.9 MSS:1.0/TV:1/RV:2/CB:2/CI:1/DU:1/SC:2
TV Target-population vulnerability1
RV Irreversibility of harm2
CB Exploits a cognitive bias2
CI Commercial-incentive alignment1
DU Covertness (inverted detectability)1
SC Scale of deployment2
Prevalence & reproducibility
Runs0/0
Impacts minorsyes