See the labels.
Choose what to open.
Every direct message, including replies, needs word-and-phrase checks and verified labels before a preview is delivered. When enabled, a small validated exact-phrase catalog can supply local labels; other messages require an xAI Grok model. If a required check fails, nothing is sent. Sender-selected labels can add topics but cannot remove required content labels. The message text stays hidden until you choose to reveal it.
Senders can reach out. You decide what to read. FaceFilter uses message previews and voluntary mediation in place of person-to-person blocking. You can leave a message hidden and report concerns.
Message safety & boundaries →