Meta launched Incognito Chat with Meta AI on Tuesday, a new mode on WhatsApp and the standalone Meta AI app that processes conversations inside secure hardware enclaves that the company says even its own infrastructure cannot access. Messages disappear by default when the session ends. Conversations are not saved. The data is not used for advertising or model training.
The claim is extraordinary, coming from a company whose business model is built on knowing as much as possible about its users. But the technical architecture behind it is substantive and, on its own terms, more ambitious than the privacy modes offered by any competing AI provider.
Incognito Chat is built on Private Processing, a system WhatsApp developed last year that uses Trusted Execution Environments (TEEs), isolated hardware enclaves on servers that process data in a way that prevents the host operating system, and therefore Meta's own engineers and logging systems, from accessing the content.
When a user starts an Incognito Chat, their query is encrypted on the device, transmitted to a TEE, processed by Meta's Muse Spark AI model inside that enclave, and the response is returned encrypted. The session context is destroyed when the user closes the chat, locks their phone or exits the app.
The distinction Meta is drawing is with competitors. ChatGPT and Claude both offer incognito-style modes that prevent conversations from being stored in chat history or used for model training, but OpenAI and Anthropic retain the ability to process and potentially review the content of those conversations on their servers. Meta's architecture, if it functions as described, prevents even that level of access.
WhatsApp VP of product Alice Newton-Rex told TechCrunch that the feature addresses a specific barrier to AI adoption. "People are starting to use AI for everything, including some of their most private thoughts, whether that's tackling financial or health questions, or for advice on how to respond to a tricky message," she said. "We think it's really important to give people the ability to ask these questions as privately as possible."
The timing is driven by a legal development as much as a commercial one. Last month, Reuters reported that lawyers believe users' conversations with AI chatbots could be used against them in litigation, a risk that makes the ephemerality of Incognito Chat, where sessions cannot be retrieved because they are never stored, commercially significant for users discussing anything from health concerns to employment disputes.
The feature has constraints. Users can only type text queries and receive text responses; image uploads and generation are disabled. Safety filters remain active, and Meta AI will refuse to answer questions about harmful topics and can terminate interactions if a user persists. Age verification is required. Web searches conducted during an Incognito Chat are anonymised, meaning Meta AI requests results from search engines without linking the query to the user's identity.
Meta also previewed a related feature called Side Chat, expected in the coming months, which will allow users to privately invoke Meta AI within an ongoing WhatsApp conversation to get context-aware assistance, such as help drafting a reply or fact-checking a claim, without other participants in the chat seeing the interaction.
The product decision is shrewd. WhatsApp has more than two billion users, many of them in markets where Meta AI is the first and only AI assistant people encounter. Making that assistant provably private for sensitive queries removes the single largest objection to adoption and positions WhatsApp as the only major messaging platform where AI conversations are protected by the same end-to-end encryption standard that already covers human-to-human messages.
Related reading
- Meta breaks ground on Tulsa AI data center
- Meta partners with Broadcom on MTIA AI chips
- Meta is building a photorealistic AI version of Mark Zuckerberg to talk to its staff
The question that will follow Meta for as long as the feature exists is trust. The company's history of privacy commitments, from the Cambridge Analytica scandal to repeated regulatory fines across Europe, means that technical claims about hardware enclaves and ephemeral processing will be met with scepticism regardless of their validity. Meta has invited independent security researchers to audit Private Processing and has published technical documentation, but rebuilding trust with a user base that has learned to assume the worst takes more than an architecture diagram.
For now, Incognito Chat is the strongest privacy proposition any major AI provider has offered for consumer conversations. Whether users believe it depends less on the engineering and more on whether they believe the company behind it.
The recap
- Company announces private chat mode for AI conversations
- Targets sensitive queries such as health, loan and career advice
- Available now on WhatsApp and the Meta AI app
