OWASP Global AppSec: How to hide prompt injection in plain sight

Words embedded in images can often bypass LLM filters that screen out harmful text instructions.