Initial detection on the device
Email addresses, phone numbers, and card numbers are detected within the Chrome extension.
LEAK GUARD
Generative AI data leak prevention | Chrome extension for organizations
Keep using AI.
Stop accidental disclosure before sending.
Checks run just before sending and display options such as cancel or send once, depending on the result.
PRODUCT PROMISE
Type without interruption. Get protection when you send.
Warn or block based on the result
Contact information prompts a confirmation, while high-risk information cannot be sent in its original form.
Record results by organization
Detected information is masked, and warnings, blocks, and usage outcomes are reflected in the admin console.
PROTECTION FLOW
From send action to inspection, decision, and record.
- 01
Intercept sending
Intercept the Enter key or send button just before submission.
- 02
Run initial detection on-device
Detect email addresses, phone numbers, and card numbers within the Chrome extension.
- 03
Optional isolated advanced detection
Administrators choose whether to enable it. Only when enabled is raw text encrypted and temporarily inspected in an isolated environment.
- 04
Warn or block
Guide the user to cancel, confirm, or take other actions based on the result.
- 05
Record the result
Reflect masked information and the final outcome in the admin console.
Detected information is masked automatically. A BLOCK result cannot be overridden to send the original text containing the detected information.
DETECTION & RESPONSE
Block high-risk data. Warn on contact details.
Pre-send behavior depends on the information type and risk.
API keys and tokens
Block sending the original text containing the detected information.
Credit card numbers
Block sending the original text containing the detected information.
Japanese individual identification numbers
Block sending the original text containing the detected information.
Email addresses
Users can cancel or review the content and choose to send it just this once.
Phone numbers
Users can cancel or review the content and choose to send it just this once.
Attachment filenames
Warn on suspicious names. Attachment contents are not inspected.
ADMINISTRATION & OPERATIONS
See received warnings and blocks across your organization.
Screens are under development. Counts, dates, identifiers, and other displayed values are examples.
DATA PROTECTION
Protection continues after the send check.
No raw-text storage
Email addresses, phone numbers, and card numbers are detected on the client. Even with advanced detection enabled, raw text is processed temporarily only in an encrypted, isolated environment and is never stored or logged.
Mask and encrypt
Only masked prompt text is stored, encrypted for each detection event.
Audit viewing actions
Decryption requires an explicit action, and its success or failure is audited. The displayed prompt text is discarded when navigating away.
Currently supports ChatGPT on the web in desktop Chrome. Only attachment filenames are inspected; file contents are not.
FAQ
What to know before adopting LEAK GUARD.
Which AI services and browsers are supported?
Currently, LEAK GUARD supports ChatGPT on the web in desktop Chrome. Ask us about other browsers and generative AI services when discussing implementation.
What is the difference between WARN and BLOCK?
With WARN, users can cancel or review the content and send it just this once. With BLOCK, the original text containing the detected information cannot be sent.
Are my prompts sent to a server?
Email addresses, phone numbers, and card numbers are detected on the client. Only when advanced detection is enabled is raw text encrypted and sent to an isolated environment, where it is processed temporarily without being stored or logged. The admin console records masked prompt text, detection decisions, final outcomes, and related information.
Can I prevent raw text from being sent to the isolated environment?
Yes. Administrators can disable advanced detection. Raw text is then not sent to the isolated environment, and only client-side detection of email addresses, phone numbers, and card numbers runs. Information covered by advanced detection is not inspected in this mode.
Are attachment contents inspected?
Attachment filenames are inspected. File contents are not.
Does the admin console always display prompt text?
The list displays decisions and metadata without decrypting the prompt body. An explicit action is required to view masked text when needed, and the result of that action is audited.
How does implementation work?
Through a paid proof of concept for organizations, we define departments, devices, and operating rules, set up administrators and license seats, and then validate WARN, BLOCK, and the admin console in real operations.
PAID PROOF OF CONCEPT
Validate your workflow with a paid organizational pilot.
- 01
Define the scope
Confirm departments, devices, and operating rules
- 02
Set up administrators
Prepare admin accounts and license seats
- 03
Activate devices
Enter activation codes in the Chrome extension
- 04
Validate in real operations
Check WARN, BLOCK, and the admin console
CONTACT