Prompt Guard
Review what you share with Prompt Guard. Check your ChatGPT prompt locally for sensitive details, edit the draft, and decide what to send.
Watch a draft with contact details become a message ready to share.
This automatic demo uses fictional contact details and a sample reply. It does not send anything to ChatGPT.
A moment to review before you send
Keep sensitive details out of an everyday prompt
Notice the details that came along with your notes
An interview follow-up needs the customer’s feedback, but may not need their email address or phone number. In the demo, a draft includes both. When the cursor presses Send, prompt guard checks the text locally and pauses the submission to show what it found.
Review the excerpt, then make your choice
The review window groups findings by category and shows each value in context. Choose Edit message to return to the original draft. The demo then replaces the sample contacts with placeholders and submits the revised message, which is checked again. The edits are made in the composer; prompt guard does not redact text automatically.
Turn it on once in the Toolbox
With Power access, open StylerGPT and enable Prompt Guard in Toolbox. It starts turned off and works at submission, so you can compose your draft without a warning on every keystroke. If you intend to share a flagged value, Send anyway lets you continue explicitly.
Keep the limits in view
The local check looks for supported patterns in outgoing text. It can miss information, and attachments, images, voice and connector uploads are outside its scope. If a check cannot finish, you can edit the message, try again or explicitly send without checking. Review your own text before you share it.
About Prompt Guard
How do I turn on Prompt Guard in ChatGPT?
Prompt Guard requires Power access and starts turned off. Open StylerGPT, go to Toolbox and enable Prompt Guard. It checks supported outgoing text when you submit a message; there is no separate Settings dialog to configure. In this demo, the pinned extension icon opens a simulated StylerGPT panel.
What sensitive data can Prompt Guard detect?
Prompt Guard looks for supported patterns such as email addresses, phone numbers, payment card numbers, bank account details, tax identifiers, API keys and secrets, login credentials and private keys. Detection is based on local rules. It can miss sensitive information or flag harmless text, so always review what you intend to share.
What happens when Prompt Guard finds something?
Sending pauses and a review window lists the detected categories and text excerpts. Choose Edit message to return to your unchanged draft, or Send anyway to explicitly continue with that message. Closing the window or pressing Escape also returns you to editing. Prompt Guard does not automatically remove or replace the detected text.
Does Prompt Guard send my prompt to another service?
The check runs locally in the extension. Prompt Guard does not send the prompt or findings to an external analysis service or store them as an analysis history. If you proceed with sending, your message still goes to ChatGPT as usual. Other enabled tools, such as prompt history, have their own storage behavior.
Does it check attachments, images or voice messages?
No. Prompt Guard checks supported outgoing prompt text. Files, images, voice and connector uploads are not checked. It is a helpful review step, not a guarantee that a message contains no sensitive information.
What if the check cannot finish?
You see Prompt could not be checked, with Edit message, Try again and Send without checking. Sending waits for your choice. Very long or unsupported drafts can lead to an incomplete check; the current text limit is 100,000 characters. A completed check with no findings allows the message to continue without a review window.
Give every prompt a moment of review
Add a local check before sending and stay in control of what you share with StylerGPT.