SafeReplies

SafeReplies reviews the subject and body of the active Gmail compose draft after the user pauses typing. A small status control shows when a draft is being checked, when no risks are found, and when findings need attention. When SafeReplies identifies risky wording, the compose assistant opens a compact review panel in Gmail. Each returned rule has its own ordered card showing the exact matched wording, risk type, severity, explanation, guidance, eligible Advisory checkpoint and policy source. The summary lists risk types without duplicating card explanations. Before Gmail sends a changed draft, SafeReplies performs an observational final check when possible. Gmail's native Send is never replaced, prevented or gated. SafeReplies does not rewrite, address or send email and has no rewrite, send-anyway, manager-approval or self-approval controls. SafeReplies processes the active compose subject and body, count-only attachment presence and, for a reply or forward, the relevant prior-message text visible in that active compose context. It does not request Gmail API scopes or scan mailbox-wide inbox or sent history, unrelated messages, other or hidden drafts, recipients, thread IDs, attachment names or attachment contents.