Meta launches end-to-end encrypted AI chats on WhatsApp using Private Processing technology

Meta is introducing end-to-end encrypted AI conversations on WhatsApp through a feature called Incognito Chat with Meta AI. The company claims messages sent in this mode are processed in a secure environment that neither Meta nor other parties can access.

Technical implementation

The feature is built on Meta's Private Processing technology, first announced at the company's LlamaCon conference in April 2025. According to Meta, messages in Incognito mode are not saved and disappear by default. The company states these conversations will not be used to train its AI models.

Meta distinguishes its approach from competitors by noting that while other apps claim to offer incognito AI chat modes, those services can still access both sides of the conversation. WhatsApp has offered end-to-end encryption for human-to-human chats since its early years.

Rollout timeline

Meta will gradually deploy Incognito Chat with Meta AI to all users over the coming months through both the WhatsApp app and the standalone Meta AI app. Availability timeline was not specified beyond "coming months."

Additional features in development

The company is also building Side Chat with Meta AI, a feature that will allow users to privately query the AI assistant about ongoing conversations. Users will be able to ask questions about a chat or request message summaries. This feature is also scheduled for release in the coming months.

Context and reception

The announcement follows Meta's recent introduction of a Grok-style chatbot feature on Threads, which faced immediate backlash from users who discovered they could not block the @meta.ai account on the platform.

What this means

This marks the first major implementation of end-to-end encryption for AI assistant interactions at scale. If Meta's technical claims hold up to independent verification, it addresses a significant privacy gap in AI chatbot services where sensitive health, financial, and personal queries are typically visible to the service provider. The move could pressure competitors like OpenAI, Google, and Anthropic to implement similar privacy protections, though the technical challenge of running AI inference within encrypted environments remains substantial.