Now, defenders are embracing the prompt injection, too

"Context bombing" tricks hacking agents into shutting down before they can do harm.
This is a summary curated by AIFuture. Read the complete article at the original source:
Read the full story on Ars Technica