Prompt Injection Exposure Audit for Your Own Assistant (Defensive)

Identify where untrusted text can reach your assistant's instructions.

Added October 15, 2026. Written with AI assistance and checked for clarity and placeholders; not reviewed by a lawyer, accountant, clinician or other professional. How these prompts are made

The Open buttons send the prompt text as currently shown below to that site only when you click. Fill in the placeholders first.

Audit my assistant for prompt injection exposure, defensively and for a system I own. Where text from outside the system enters (user messages, uploaded files, web pages, emails, tool outputs): [UNTRUSTED_INPUTS]. What the assistant can do with that text (tools, data access, actions): [CAPABILITIES]. Existing protections: [PROTECTIONS].

Map each input path to the capabilities it could influence and rate the exposure (low, medium, high) with the reason. Make a table with Input path, Possible effect, Safeguard in place, Gap, and Recommended control (separate instructions from data, least privilege, human approval for risky actions, output checks, logging). Count high-exposure paths. Provide a plan for testing my own system in a safe environment with benign test strings and a checklist for monitoring.

Do not provide attack payloads or techniques for use against systems I do not own.

Add a worked mapping: the assistant reads customer emails (untrusted) and can draft replies (low impact) but also can look up order details (medium) and issue refunds (high); the exposure is highest where email text can influence refunds, so require approval and separate data from instructions. Provide benign test strings for my own system.

Fill in before use

Optional: type your details and the prompt above updates. Nothing is saved or sent.

Replace each bracketed placeholder with your own details, then paste into Claude, Gemini, ChatGPT or another assistant.