Send Guard
Send Guard is Sonomos’s safety net for websites that don’t have dedicated Cloak support. It warns you — or blocks submission entirely — when PII is detected in text you’re about to send.
How it works
When you’re on a site that Cloak doesn’t natively cover, Send Guard monitors form fields and text inputs. If Dagger detects PII in content you’re about to submit, one of two things happens depending on severity and your configuration:
Banner mode
A non-blocking warning banner appears at the top of the input area, listing the detected PII categories (e.g. “Detected: 1 SSN, 2 emails, 1 name”). You can:
- Dismiss the banner and proceed (the form still submits)
- Edit your text to remove the flagged data
- Click any item in the banner to jump to its location in the input
Banner mode is intended for medium-severity matches where you probably want to know but don’t want to be interrupted.
Blocking modal
For high-severity detections (SSNs, credit card numbers, medical records), a modal overlay prevents submission. You must explicitly acknowledge the risk or remove the PII before the form will submit. From the modal you can:
- Review each detected entity individually, with the surrounding context
- Toggle entities on/off so you can mask only what’s actually sensitive
- Acknowledge & send to override and submit anyway (logged in your detection history)
- Cancel to return to the form and edit
The modal is deliberately intrusive — it’s the last line of defense before high-severity PII leaves your machine on a site Cloak can’t transparently rewrite.
When Send Guard activates
Send Guard runs on every site that isn’t covered by Cloak’s automatic masking. Currently, Cloak fully supports Claude.ai, Gemini, and Grok. All other websites — including ChatGPT, internal tools, CRMs, ticket systems, social media, email composers, and so on — get Send Guard automatically.
It activates on:
- Standard
<textarea>and<input>elements - Most rich-text editors (contenteditable surfaces)
- Common chat composers and form builders
It does not activate on:
- Browser-internal pages (
chrome://...,about:...) - Pages where the extension can’t run due to browser policy (e.g. extension stores)
- File downloads or non-text submissions
Configuration
Send Guard is enabled by default. You can adjust its behavior in the Sonomos popup under Settings → Send Guard:
| Mode | Behavior | When to use |
|---|---|---|
| Banner + Modal (default) | Banner for medium severity, modal for high severity | Recommended for most users |
| Modal always | Modal for any detection, even low severity | Highly regulated environments |
| Banner only | Warnings only, never blocks | When workflow speed trumps blocking |
| Off | Disable Send Guard entirely (not recommended) | Testing, or strictly Cloak-only flow |
On the Teams plan, admins can pin a minimum Send Guard mode so members can’t drop below it.
Tips for living with Send Guard
- Edit, don’t acknowledge. Acknowledging a high-severity warning is a fine emergency escape hatch, but the safer habit is to edit the input until the warning clears.
- Use placeholders yourself. If you find yourself acknowledging the same SSN repeatedly, replace it with
[SSN]in your own text — the AI usually responds well to placeholders. - Watch the count. The banner / modal title shows how many entities were detected. A surprisingly large number is often a clue you’ve pasted more than you meant to.
Coming in Sonomos Desktop
The same warn-and-block model, but applied OS-wide:
- Triggered in any app — native AI clients, IDEs, chat apps, document editors.
- Triggered when you copy PII to the clipboard, before any other app can paste it.
- Triggered when a screenshot or dragged file contains PII headed for an AI app.
- Per-app rules, so your EHR doesn’t get blocked but your public chatbot does.
See the Sonomos Desktop overview for more.