Keep company information from slipping out by accident.

Personal or confidential information can end up sent to an external AI service without anyone meaning it to. Security Filter guards that entry point.

Security Filter

Detect before it's sent. Mask it. Restore it later.

Security Filter detects PII before it reaches a model, sends it masked, and restores it on response — standard on Business and above.

How it works

Detect

Scan outgoing content for PII.

Mask

Mask what's detected.

Send

Send the masked content to the model.

Restore

Restore the masked content when the response arrives.

By plan

Available on Business, Business+ and Enterprise. On Enterprise, action choices and your own rules are enforced across the whole organization.

項目BusinessBusiness+Enterprise
Auto-mask and restoreYesYesYes
Custom detection rules, BYO dictionaries & org-wide enforcement——Yes

How each tier behaves, with an example

Say someone sends "Write a reply about Mr. Tanaka (tanaka@example.com / 090-1234-5678)" to an external AI. Here is what happens on each plan.

Business and Business+ — masked automatically, restored in the reply

The email address and phone number are replaced with placeholders such as "<Email_1>" and "<Phone_1>" before sending. The external AI does not receive the values that were masked. Values it did not find, and text shown inside an image, are not covered. If the reply contains a placeholder, it is turned back into the real value only when shown on screen, so the person keeps working as usual. Personal names such as "Mr. Tanaka" are not found out of the box.

Enterprise — choose the action, add your own rules, enforce it organization-wide

For each kind of data you can choose: mask and send, warn and send, or stop the send. Names and internal terms can be added as your own rules. The setting applies to the whole organization and individuals cannot turn it off.

A policy enforced for the whole organization

From Business upward, auto-mask and restore is built in. On Enterprise, organization admins use the admin screen to choose the action for each kind of data, add their own rules (keywords and regular expressions), and switch what is covered (conversation history, attachments, knowledge, web page text and generation prompts). Some paths check only the current message, and some check everything regardless of the switches. The setting applies to everyone, and individual users cannot turn it off. When a send is stopped, it is recorded in the audit log and nothing reaches the AI or uses PFC. See the Enterprise admin guide for setup.

Design approach

The Security Filter is designed to coexist with perimeter DLP (Data Loss Prevention), not replace it. Perimeter tools look at all traffic; the Security Filter looks at the content being sent to an AI and detects and masks it before sending.

Your own rules, for terms only your company uses, are available on Enterprise. It finds things that fit fixed patterns and the rules your organization adds, so it does not guarantee that every piece of sensitive information is found. The contents of images, audio and video (including text shown in an image) are not checked.

Try three modes — Chat, Compare, and Debate — free.

Start free
Language日本語English