Researchers from Zenity have found multiple ways to inject rogue prompts into agents from mainstream vendors to extract ...