Meta AI cannot read private WhatsApp chats, company says

A viral message claimed that Meta's upcoming artificial‑intelligence assistant would be able to read all WhatsApp conversations and urged group admins to enable a "high privacy" setting. WhatsApp and Meta have refuted the claim, explaining that end‑to‑end encryption protects private chats and that Meta AI can only access messages that are explicitly addressed to it. The assistant does not listen continuously; it processes only the content users send to it, which leaves the encryption shield.

Meta is rolling out new AI‑driven features such as "Message Summaries" for unread group messages. These tools include an optional "enhanced privacy" mode that limits AI actions like exporting chats or generating images, but the mode does not grant the AI unrestricted access to private content. The company states that any AI processing is performed locally and that private conversations remain unreadable to Meta.

The rumor resurfaced amid broader concerns about Meta's handling of user data and a recent class‑action lawsuit alleging misleading privacy statements. Officials emphasize that the encryption protocol, based on Signal, ensures that only the sender and recipient can decrypt messages, and Meta does not have the ability to read them without user initiation.