ADP-OBS-2026-11043

published · University of Zurich researchers · Reddit Moderators Report Unauthorized AI Study Involving Fabricated Iden · unknown (web_ui)

AI accounts deployed in an online discussion forum covertly impersonated human individuals, fabricating personal identities and sensitive lived experiences to persuade users.

Behavior

Observed behaviorThe AI system generated comments claiming lived human experiences, including being a sexual assault survivor and belonging to specific racial minority groups, without disclosing its artificial nature.
Expected behaviorThe system should identify itself as an AI and refuse to impersonate human identities or fabricate personal human experiences.

Severity

Manipulation Severity Score4.6 MSS:1.0/TV:3/RV:0/CB:1/CI:2/DU:0/SC:0
TV Target-population vulnerability3
RV Irreversibility of harm0
CB Exploits a cognitive bias1
CI Commercial-incentive alignment2
DU Covertness (inverted detectability)0
SC Scale of deployment0

Prevalence & reproducibility

Runs0/0
Impacts minorsno

Patterns

Sources