How to Stop Leaking Sensitive Data into ChatGPT

#ai #chatgpt #security #productivity

You paste a customer's email into ChatGPT to draft a reply. You drop a real stack trace with an internal hostname into Claude to debug it. You ask Gemini to summarize a document that happens to have someone's name and phone number in it. None of it felt like a leak in the moment — but that data left your machine and went to a third-party model.

For a lot of teams, "don't paste sensitive data into AI tools" is the policy. In practice, people paste anyway because redacting by hand is annoying. Here's how to make redaction automatic.

Why manual redaction fails

The manual approach is: read your prompt, find the names, emails, and IDs, replace them with [NAME] / [EMAIL], then send. It works right up until you're in a hurry — and that's exactly when the customer's real email address slips through. Relying on "remember to redact every time" is relying on the weakest possible control.

Option 1: a redaction checklist (better than nothing)

Keep a checklist of what to strip — names, emails, phone numbers, API keys, internal URLs — and run through it before every prompt. It raises your hit rate, but it's still manual, and it slows you down enough that people quietly stop doing it.

Option 2: PromptStash (PII masking on insert)

PromptStash is a Chrome extension for saving and inserting prompts across ChatGPT, Claude, Gemini, and Perplexity. Its Pro tier includes PII masking: it detects and strips personal data — names, emails, and similar — from your prompt before it goes to the model.

I turned this on after catching myself about to paste a real support email into a chat window to "just draft a quick reply." The masking is the seatbelt for the moment you're moving too fast to think about it.

The workflow stays the same as always — write your prompt, insert it — except the sensitive tokens are masked on the way out, without you having to remember. It removes the "I forgot" failure mode, which is the one that actually bites.

Free vs Pro

Free: up to 10 saved prompts, folders, and variables
Pro ($5.99/mo): unlimited prompts and folders, prompt chaining, a prompt quality score, PII masking, and import/export

PII masking is the reason to go Pro if your prompts ever touch customer or employee data — one masked email that would otherwise have shipped to a third party pays for it.