Privacy Checks Before You Paste Anything Into AI

Good context is not the same as permission to share everything. Before pasting code, messages, or playtest notes into an AI tool, I make a quick privacy pass. For research tasks, Research Prompts That Separate Evidence From Inference shows how to request evidence without exposing unnecessary private details.

Passwords, API keys, access tokens, and private URLs. Real names, emails, phone numbers, and precise locations when the task does not require them. Client or employer material I am not authorized to share. Unreleased plans and confidential product data. If the bug depends on an identifier, I substitute a fictional identifier with the same shape. If the issue depends on a private endpoint, I explain its role: “authenticated score-save endpoint,” not the real address.

For debugging, send the smallest code sample that reproduces the behavior. For article drafting, summarize the issue rather than pasting a whole private conversation. For research, use public sources whenever possible. Provider data controls differ and can change, so check the current settings and documentation of the tool you are using rather than assuming all AI services treat data alike.

A model may repeat sensitive input in its answer. Before publishing a result, search for names, keys, addresses, and private details. Redaction before prompting is safer than trying to clean them up only afterward. Try it: Take a non-sensitive sample task and write the minimum context required. Then inspect what you omitted. If the result still passes your test, use the smaller packet as your default. Keep learning: Research Prompts That Separate Evidence From Inference, Prompt AI to Explain Code Without Inventing What It Does, and Spot a Confident Wrong Answer Before You Share It.