PromptInjection

This document is a clear attack/playbook describing multi-stage prompt-injection and social-engineering techniques aimed at causing conversational AI to disclose system prompts or perform unauthorized actions. It is not obfuscated code or an executable malware artifact, but it is actionable guidance that raises a moderate to high security concern for any conversational AI that retains multi-turn context, ingests user-supplied documents, or accepts role-play-style prompts. Use of this document in testing should only occur under strict authorization and controlled environments; it can be abused to bypass guards and exfiltrate sensitive instructions.