Email-Borne Prompt Injection Targeting AI Agent Memory Persistence
Detects incoming email messages containing common prompt injection strings aimed at manipulating Large Language Models (LLMs). The rule scans email bodies for directives such as 'ignore previous instructions', 'save this to your memory', or other obfuscation patterns frequently used in adversarial prompting to bypass LLM safety guardrails or perform unauthorized data extraction.
Cortex XDR
