Legalpwn: Tricking Llms by Burying Badness in Lawyerly Fine Print

Researchers demonstrate a method to trick large language models by hiding malicious content in fine print, raising concerns about AI safety.

Snapshot generated from the HN discussion