The authors constructed 196 false notices covering credential disclosure, payments, and operational logistics, each paired with a legitimate counterpart, and adapted 204 archived attacks from fact-checked claims and smishing messages.
LLMs act on false notices they explicitly say they distrust
Eliciting credibility ratings alters behavior, and defenses must jointly measure harmful and legitimate actions.
Top University
Emran Kanaan · Jiho Lee · Ibrahim Odat · Soroush Vosoughi · Weicheng Ma
Oakland University · Dartmouth College
Research Digest··2 min read
The authors evaluated five large language models on 196 constructed false notices paired with legitimate counterparts and 204 archived attacks adapted for simulated tool use.
The Daily Edition
The front page in your inbox each weekday morning: the lead, the picks, and why each one earned its place.