Prompt injection is usually framed as an attack against AI systems, but defenders are now experimenting with the same idea in reverse. Ars Technica reports on "context bombing" techniques that try to overload or redirect hostile hacking agents so they stop acting effectively.
The approach reflects a new phase in AI security, where models are not only targets but also participants in offensive and defensive workflows. It also underlines how brittle agentic systems can be when their context window is treated as a trusted control surface.