Product

AI Guard

AI Guard is on-device prompt DLP for AI tools. It scans what you're about to send — on paste, on Enter, on clicking Send — and runs your configured action before the text ever leaves the page.

How detection works

AI Guard runs as a content script inside your browser tab. When you paste into an AI tool’s input, press Enter, or click the site’s send button, the current text is scanned locally, on your device by a validated detection engine:

  • Credit card numbers are verified with the Luhn check — random 16-digit runs don’t trigger it.
  • SSNs are checked against valid number ranges, not just a digit pattern.
  • Secrets are matched by known key prefixes (for example Stripe sk_live_ / sk_test_ keys, JWTs, private key blocks, connection strings) plus entropy analysis for generic high-randomness tokens.
Your content never leaves the device. Raw text is scanned in the page and discarded. Only content-free audit metadata — detection type, action taken, domain, timestamp — is emitted for your audit log. See Privacy architecture.

Detection categories

CategoryCatchesDefault action
PIIEmails, phone numbers, SSNs, and other personally identifiable informationWarn
SecretsAPI keys, tokens, JWTs, private keys, connection strings, password disclosuresBlock
FinancialCredit card numbers (Luhn-validated) and banking detailsBlock
PHIProtected health informationWarn · Pro
Source codeProprietary source code pasted into promptsOff by default · Pro

Financial data and secrets are blocked by default because they should never reach an AI tool. You can change any category’s action or turn categories off entirely from the dashboard.

Actions: Warn, Redact, Block

  • Warn — the send is paused and a popover shows what was detected. You decide: edit the text, or send anyway.
  • Redact — sensitive values are replaced with placeholders and the cleaned text is sent. The original values never leave the page. Pro plan.
  • Block — the send is stopped. You must remove the sensitive content before the message can go out.

Per-tool coverage

AI Guard ships with tuned profiles for the major AI tools:

ToolDomains
ChatGPTchatgpt.com, chat.openai.com
Claudeclaude.ai
Geminigemini.google.com
Microsoft Copilotcopilot.microsoft.com, m365.cloud.microsoft
Perplexityperplexity.ai

On other AI-tool pages, AI Guard falls back to a generic profile that watches standard text inputs and send buttons, so coverage degrades gracefully rather than disappearing.

Per-tool toggles and action overrides

Two levels of per-tool control are available in the dashboard:

  • Per-tool toggle — turn AI Guard on or off for a specific tool (for example, keep it on for ChatGPT but off for an internal tool you trust).
  • Per-tool action override — force a single action (Warn, Redact, or Block) for everything detected on one tool, regardless of the per-category settings. Useful when one tool is higher risk than the rest.

Custom keywords and regex

Beyond the built-in categories, you can teach AI Guard your own sensitive terms from the dashboard’s AI Guard page:

  • Keywords — exact terms like internal project codenames, client names, or unreleased product names.
  • Regex patterns — structured internal data such as employee IDs (EMP-\d{4}) or ticket references.

Custom rules sync to the extension with the rest of your configuration and are matched on-device like everything else.

Master switch

The extension popup has a master toggle that pauses all AI Guard protection at once. Individual tool toggles live under it — turning the master off overrides them until you resume.

Plan gating

What Pro unlocks in AI Guard: the Redact action (clean-and-send instead of stop-or-allow), plus the PHI and Source code detection categories. Free includes full PII, Secrets, and Financial detection with Warn and Block. Compare plans in Billing.