WhatsApp launches Incognito Chat with Meta AI, promises fully private conversations

WhatsApp is launching Incognito Chat, a mode for talking to Meta AI that the company calls fully private. According to Meta, not even the platform itself can read the prompts and responses: they are processed in a secure environment and vanish when the session ends.

How it works

The new mode is built on Private Processing, the technology WhatsApp introduced a year ago for AI features such as message summaries and writing suggestions. The idea is to give users access to large models without breaking the messenger's familiar promise of private conversations. Instead of ordinary server-side processing, requests go to an isolated execution environment where they cannot be viewed, even from inside Meta's infrastructure. In essence, Meta is trying to reproduce the properties of an end-to-end approach for tasks that cannot be run entirely on a smartphone.

WhatsApp head Will Cathcart compared it to a giant phone for AI that has no password, even for the company itself. Meta says that under this setup it sees only the fact that the feature is being used, not the content of the conversation. Incognito Chat began rolling out on May 13, 2026, in WhatsApp and the Meta AI app, and the global rollout will take several months.

What users get

In interface terms, it is a separate temporary session with Meta AI, meant for questions people usually do not want in their regular chat history: from finance to career advice and personal problems. Meta is betting that a private mode will lower the main barrier to using AI in a messenger.

• The conversation is visible only to the user
• Messages are not saved by default and disappear after you leave the session
• The mode currently works only with text
• For questions about recent topics, the AI can anonymously use web search; this option can be turned off
• In the coming months Meta promises Side Chat: private AI suggestions about the current chat without interfering with the main conversation

Meta also says it will later add image processing and voice recognition. If these arrive within the same Private Processing perimeter, the company could expand use cases from simple questions to analyzing photos, voice messages and group discussions. For WhatsApp this is an important test: users will either accept AI as a built-in tool or conclude once and for all that an assistant has no place in private messaging.

Where questions remain

Meta promises external audits, a vulnerability reporting program and code review by independent specialists. This matters because the whole design rests not only on cryptography but also on trust in how the company deploys and maintains the system in the cloud.

Johns Hopkins cryptographer Matt Green, who is familiar with the Private Processing architecture, believes the idea is workable overall.

"If you want to talk to an AI so that no one sees the conversation, including

Meta, this scheme should work."

But even with such an architecture, the risk does not go away. Any secure cloud system remains an attractive target for attacks, and the cost of a mistake is especially high here: users may share medical, financial and family details with the bot.

The broader backdrop around Meta adds to the skepticism. The same week, the company fully removed opt-in end-to-end encryption from Instagram Direct, so its approach to privacy does not look equally strict across all products.

What this means

If Incognito Chat works the way Meta promises, it could become the first truly mass-market example of a private AI chat inside a messenger. For the market, it signals that the next competition will be not only over model quality but over who can more convincingly prove that your AI requests will not sit on someone else's server in readable form.

*Meta has been designated an extremist organization and is banned in Russia.