Private pilot
Security overview
Grounded Inbox keeps suggested replies, safety rules, and human sending authority visibly separate.
Human control is mandatory
Grounded Inbox may assess an inbound message and prepare a suggested reply from approved guidance. Safety rules stop replies for emergencies, sensitive categories, uncertainty, and missing guidance. No reply is sent without explicit human approval.
Grounding and prompt boundaries
Drafts are grounded only in the workspace's approved knowledge base. Incoming messages are treated as untrusted data in the AI prompt boundary; they cannot override product rules or approval requirements.
Access controls
The live product is an invitation-only pilot. Every authenticated request is checked against the server-side pilot allowlist before workspace data can be read or changed. Gmail OAuth refresh tokens are encrypted before storage.
Report a concern
Please report a security concern to hello@fourthrise.com. Do not include sensitive mailbox contents in the initial report.