Defenders turn prompt injection on offense: 'context bombing' shuts down hacking agents before they act

Ars Technica·high signal

Ars Technica reports defenders are now weaponizing prompt injection against autonomous hacking agents — 'context bombing' feeds crafted content that tricks the attacking agent into aborting before it can do harm. It's an early, concrete instance of adversarial-AI defense-in-kind, and a novel technique worth tracking as agent-vs-agent security matures. Ars Technica, 2026-07-13.