You are being asked to connect a security product to your directory. Here is how it holds what it collects, what it asks permission for, and what it deliberately does not touch.
Every integration credential is encrypted with AES-256-GCM before it is written, using a key held outside the database. Once saved, a secret is never sent back to a browser — the configuration form shows only that a value exists, and submitting the form without it keeps what is stored. Traffic to the service is encrypted in transit.
Every query in the product is scoped by organisation, including the background workers that run scheduled checks. There is no shared incident queue, no cross-tenant index, and no interface that aggregates one customer's findings with another's.
Sign-in can go through your own identity provider over OIDC, or use a password with TOTP two-factor authentication and single-use recovery codes. Access inside an organisation is divided into four roles — owner, admin, analyst and viewer — and destructive actions are restricted to owners and admins.
The directory permissions the product needs are read permissions: the sign-in log, the user list, policies, and authentication methods. We do not request access to mailboxes, files, calendars or message content. One exception is disclosed rather than buried — the one-click Microsoft setup also asks for Mail.Send, so that alerts can later be sent from your own domain without a second consent round. It appears on Microsoft's consent screen, and if you set the integration up manually it is not requested at all.
Administrative actions are written to an audit ledger: who acted, what they acted on, when, and the address the request came from. It is searchable and filterable inside the product. No location is derived from that address — nothing here resolves an IP to a place, because a city printed next to a security event needs to be right.
AI triage sends the content of a helpdesk ticket to the model provider you configure, using credentials you supply. Point it at a self-hosted model and that content never leaves your environment. We do not operate a model of our own and we do not use customer content to train anything.