2024-08-07 23:01:15
KoreLogic
PUBLISHED
Attackers can craft a malicious prompt that coerces the language model into executing arbitrary JavaScript in the context of the web page.