Defenders are now utilizing prompt injections, a tactic previously employed by attackers to manipulate large language models (LLMs) into executing malicious commands. By embedding specific prompts alongside sensitive data, such as passwords and cryptographic keys, researchers from Tracebit have discovered a potential method to enhance security. This approach can be used to detect and prevent unauthorized access to sensitive information, as the LLMs can be instructed to respond in a specific manner when encountering the embedded prompts. For instance, a well-crafted prompt can trigger an LLM to alert defenders when an attacker attempts to exfiltrate sensitive data1. This development highlights the evolving nature of LLM security, where the same techniques used by attackers can be repurposed for defensive purposes. The ability to leverage prompt injections for security purposes matters to practitioners, as it offers a new layer of protection against emerging threats, particularly those related to LLMs.