Defenders turn prompt injection into a defensive tool

Ars Technica reports that security researchers are now using prompt injection defensively to combat adversarial attacks on LLMs. The article details how embedding deceptive commands can help detect and neutralize malicious prompts.