“Context bombing” tricks malicious AI agents into shutting down before they can do harm.
WI
Source
This article was originally published by Wired and written by Dan Goodin, Ars Technica.
Read original article on Wired