Meta has announced “Incognito Chat with Meta AI,” a new private AI chat mode for WhatsApp and the Meta AI app.
The feature is built on the firm’s existing “Private Processing” infrastructure and is designed for sensitive AI interactions involving personal, financial, health, or work-related information.
According to Meta, Incognito Chat creates temporary AI conversations processed inside isolated secure environments, with chats disappearing by default and no conversation history retained by the service.
Meta said the feature addresses growing concerns over how AI companies handle prompts and chat histories, especially as users increasingly rely on AI assistants for private or confidential tasks. Unlike typical “incognito” or “temporary” chat modes offered by some AI platforms, Meta claims its implementation prevents the provider itself from reading prompts or generated responses.
The new mode is powered by Meta’s “Private Processing” architecture, a system the company first detailed in a technical white paper published last year. The infrastructure relies on Trusted Execution Environments (TEEs), confidential computing hardware from AMD and NVIDIA, remote attestation, and encrypted routing mechanisms intended to isolate user data from Meta’s internal systems.
Meta describes Private Processing as a server-side system designed so that “sharing messages with Private Processing does not make them available to Meta, WhatsApp, or anyone else.” The platform was originally developed for WhatsApp AI features such as message summarization and writing suggestions.
Requests sent through Private Processing are encrypted end-to-end between the user’s device and the processing environment, with traffic routed through third-party relays using Oblivious HTTP (OHTTP) to obscure users’ IP addresses from Meta infrastructure. The system also uses anonymous credentials, allowing requests to be authenticated without directly identifying the user.
Meta’s white paper says the infrastructure uses AMD SEV-SNP confidential virtual machines and NVIDIA H100 GPUs running in confidential computing mode to protect data while it is being processed. The company also implemented remote attestation mechanisms that allow client devices to verify they are connecting to approved processing environments running untampered software.
Meta says the system is stateless by design, meaning conversations are not persistently stored inside the processing environment. To support multi-turn AI chats, conversation context is instead sent from the user’s device with each request rather than being retained server-side.
Meta also announced an upcoming “Side Chat” feature that will use the same Private Processing infrastructure. The tool will reportedly provide AI assistance within existing chats while preserving the privacy protections offered by the architecture.
Incognito Chat with Meta AI will begin rolling out over the coming months on WhatsApp and the Meta AI app. Users handling sensitive information should remain cautious when sharing confidential data with any AI system, especially during early deployments, and ensure their apps remain up to date.
If you liked this article, be sure to follow us on X/Twitter and also LinkedIn for more exclusive content.
