The new mode runs Meta AI on WhatsApp inside the company’s Private Processing enclave, with conversations deleted by default and no server-side record retained.
Meta has launched an Incognito Chat mode for Meta AI on WhatsApp and the Meta AI app, an effort to address the awkward fact that its assistant, like every other major AI chatbot, has until now been able to read the conversations users have with it.
The new mode, the company announced on Tuesday, processes user messages inside what Meta describes as a secure environment that even Meta cannot see, with conversations deleted by default once the session ends.
The technical foundation is WhatsApp’s Private Processing system, the architecture the company published in April 2025 to let AI features run on encrypted data inside Trusted Execution Environments on Meta’s servers.
Inside that enclave, the model can read and respond to a query, but the contents are not accessible to Meta’s engineers, its logging systems, or any of its commercial pipelines.
Other apps offer what they call incognito modes for AI conversations, but Meta’s framing in the announcement is pointed: “they can still see the questions coming in and the answers going out.”
The launch responds directly to a category-wide privacy concern. AI chatbots have become a default tool for the sort of question users would once have asked a doctor, a lawyer, or a partner, with all the data exposure that implies.
OpenAI, Google, and Anthropic each store conversation histories by default, with varying user controls. Apple Intelligence routes some queries through Apple’s Private Cloud Compute, an enclave architecture that is the clearest existing analogue to what Meta is now shipping inside WhatsApp.
Two product details follow from the design. First, the conversations are not saved server-side at all; users cannot pull up Incognito Chat history later because there is nothing to pull up.
Second, the disappearing-by-default behaviour means even a compromised device leaks less, since the chat residue clears between sessions.
Meta has published a technical whitepaper describing the cryptographic architecture for outside review.
A second feature is on the way. Sidechat with Meta AI, also protected by Private Processing, will let users get AI help inside an existing WhatsApp conversation, with the assistant aware of the chat’s context but its responses kept invisible to the other participants.
Meta said Sidechat will arrive on WhatsApp “in the coming months,” without a firmer date.
The launch’s commercial logic is straightforward. WhatsApp has been built for a decade around end-to-end encryption as a selling point, and Meta’s pitch for AI on the platform has had to find a way around the central tension that a conversational AI assistant needs to read your messages to be useful.
Private Processing is the company’s attempt at squaring that circle. The Incognito Chat product is the first time the architecture has been put behind a user-facing feature on this scale.
Whether the implementation holds under scrutiny is a separate question. Trusted Execution Environment-based AI systems have been audited and criticised across the industry, with researchers periodically demonstrating side-channel attacks against similar architectures from Apple, Google, and the hyperscalers.
Meta has invited external review of its Private Processing design, and the new whitepaper extends that posture, but the model’s resistance to subpoena, in particular, has not yet been tested in court.
Incognito Chat with Meta AI begins rolling out on WhatsApp and the Meta AI app this week, with broader availability over the coming months.
The launch lands at the end of a difficult fortnight for Meta on the privacy front, with US employees protesting the company’s new mouse-tracking software on Monday and the company a week out from layoffs of roughly 8,000 staff.
Inside Meta, the bet appears to be that consumer-facing privacy moves like this one will outweigh the internal-surveillance optics.
