Prompt Injection Attacks Are Thwarting AI Hacking Agents Wired By: Dan GoodinοΌ Ars Technica 18 July 2026 at 05:00 βContext bombingβ tricks malicious AI agents into shutting down before they can do harm.