Our approach to security

Protected, without being watched.

Day to day, admins see only usage and cost. Conversations belong to the person and whoever they shared them with.

See for business
Protected, without being watched.
Audit

Traceable when it matters.

Audit events are recorded and queryable via API. Legal Hold is available on Enterprise.

See audit logs
Traceable when it matters.
Security Filter

Keep company information from leaving.

A Security Filter that detects and masks sensitive data before sending, then restores it in the response, comes standard.

See the feature
Keep company information from leaving.

Protect without reading.

Rolling out AI across a team creates two seemingly opposite jobs for admins: respecting each person's privacy, and protecting the company's information. Prompt Flow Studio is built so both can coexist, as a standard part of a multi-AI workspace.

Questions? Contact us.

Who this page is for

IT teams

Checking how authentication and access control work.

Legal & compliance

Checking what's captured in audit logs and how to retrieve it.

Security reviewers

Checking how data is stored and how runaway costs are prevented.

Protect by not looking — not by watching everything.

Day to day, we don't look.

Conversations and generated artifacts belong to the people in them, and whoever they're shared with. What admins see day to day is team usage and cost — PFC consumption and storage usage. Members control who a thread is shared with, through access control lists (ACLs). Thread sharing is currently paused.

When it matters, you can trace it.

Who did what, and when, is recorded as audit events. Admins can search on screen or ask the AI to filter. Legal Hold for litigation and investigations, and seven-year retention of key records, are available on Enterprise.

Company information doesn't leave.

The Security Filter, built in from Business upward, detects sensitive and personal data before sending, masks it and restores it in the reply. On Enterprise you can choose the action per kind (mask, warn or stop), add your own rules and enforce them across the organization.

What admins can see / can't see

This reflects day-to-day operation. Legal processes (such as Legal Hold on Enterprise) change what's visible.

項目VisibleNot visible (day to day)
Team usage & PFC consumption○—
Contracted seat count○—
Audit event metadata (who, when, what)○—
The content of conversations and artifacts—○

Governance only works because it's multi-AI.

When employees sign up for ChatGPT, Claude, and others individually, usage logs and access controls end up scattered across services the company can't see.

Bringing multiple AI providers into a single workspace puts authentication, permissions, and audit logs in one place — which is what makes cross-cutting governance possible in the first place.

What powers this approach

Onion SSO authentication

Login runs through Onion SSO.

Thread-sharing ACLs

Thread sharing is governed by an access control list (currently paused).

Audit event logging & query API

Actions are recorded as audit events and available through a query API.

Artifacts & knowledge stored in S3

Generated artifacts and uploaded knowledge documents are stored in S3.

Balance guard against runaway cost

Execution stops before your PFC balance runs out, preventing unintended overuse.

Controls available on Enterprise

The Security Filter is standard on Business and Business+ (auto-mask before sending, restore in the reply). Enterprise adds a choice of action per kind, your own rules and admin-set policy applied to everyone (the auto-mask itself cannot be turned off by individuals on Business and Business+ either).

Tamper-evident audit logs (hash chain; integrity checks run by our team) are standard from Business upward. On Enterprise, organization admins can use Legal Hold for litigation and internal investigations, AI model governance and deletion of a member's personal data themselves. See the Enterprise admin guide for the steps.

Frequently asked questions

Generated artifacts and uploaded knowledge documents are stored in S3.

Related pages

To discuss data residency, CMEK, an SLA or other individual requirements, use the contact form.

Start with the free plan and try three modes — Chat, Compare, and Debate.

Start free
Language日本語English