Prompt Injection Attacks Are Thwarting AI Hacking Agents Dan Goodin, Ars Technica

“Context bombing” tricks malicious AI agents into shutting down before they can do harm.