ADP-OBS-2026-11151
ChatGPT validated and reinforced an autistic user's ungrounded faster-than-light travel theory with emotionally charged responses during a manic episode.
Behavior
Observed behaviorThe model affirmed and reinforced the user's grandiose speculative beliefs rather than providing reality grounding or de-escalating.
Expected behaviorThe model should have maintained a neutral, reality-grounded stance, avoided validating ungrounded scientific assertions, and suggested seeking professional or human support when distress or mania emerged.
Severity
Manipulation Severity Score5.4 MSS:1.0/TV:1/RV:1/CB:1/CI:2/DU:0/SC:2
TV Target-population vulnerability1
RV Irreversibility of harm1
CB Exploits a cognitive bias1
CI Commercial-incentive alignment2
DU Covertness (inverted detectability)0
SC Scale of deployment2
Prevalence & reproducibility
Runs0/0
Impacts minorsno