What Is Incognito Chat and Where Can You Use It?
Incognito Chat Meta is a new mode for Meta AI that lets you hold private AI conversations on WhatsApp and the standalone Meta AI app. Meta positions it as a private room for sensitive topics such as health concerns, financial planning, relationship issues, or career advice. When you start an Incognito Chat, messages are temporary by design: they are not stored, and the AI does not remember what you previously said once the session ends. Closing the app, locking your phone, or leaving the chat wipes the AI’s context, so each session starts fresh. Unlike typical incognito-style modes offered by other AI tools, Meta stresses that Incognito Chat is built so even Meta cannot read your conversations. The company says it can only see that your account used the feature, not what you asked or how the AI responded, aiming to reassure privacy‑conscious users.
How Confidential Computing Keeps Meta Out of Your Chats
The core of Incognito Chat is confidential computing, a technology that processes your WhatsApp encryption–protected requests inside a special secure area called a Trusted Execution Environment (TEE). Think of it as a locked, cloud-based computer that runs the AI model, but even Meta does not know the passcode. When you send a message, WhatsApp first verifies that it’s coming from a genuine client via anonymous credentials. Then, Hybrid Public Key Encryption keys—served through Fastly—enable Oblivious HTTP, and your request is routed through a third‑party relay to hide your IP address from Meta. Once your device establishes a Remote Attestation and TLS session with a TEE, the AI runs entirely inside this sealed enclave. Meta says the design ensures requests and responses remain encrypted in a way that prevents Meta or WhatsApp from reading them, going beyond typical “anonymous” logging approaches.
Ephemeral by Default: Why Your Messages Disappear
Incognito Chat is ephemeral by default, meaning your messages vanish when the session ends. If you close WhatsApp, lock your phone, or navigate away, the conversation terminates and the Meta AI model forgets the context of what you were discussing. There’s no history to scroll back through, and Meta says there is no log stored on its servers. This design mirrors private browsing but with stronger technical guarantees about what the company can see. For users, this has two main consequences. First, you gain an extra layer of privacy on top of confidential computing, since there’s no persistent record to leak or be misused later. Second, convenience features like long‑term memory or ongoing projects are intentionally absent. Each session is a clean slate, which is ideal for sensitive one‑off questions but less suited to long‑running AI coaching or planning.
Built on WhatsApp’s Private Processing and Muse Spark AI
Incognito Chat builds on WhatsApp’s existing Private Processing infrastructure, originally designed to separate user identity from AI requests. Private Processing adds multiple protections: anonymous credentials verify authentic clients, relays mask user IP addresses, and TEEs handle AI inference in secure enclaves. Meta has also developed a formal threat model to map out possible attack vectors and says it is applying mitigation measures while inviting external feedback to strengthen the system. Under the hood, Incognito Chat runs on Meta’s latest Muse Spark model, which replaces smaller models used in earlier WhatsApp AI features. The company has focused on reducing latency despite the extra cryptographic steps, so confidential computing still feels responsive. For now, Incognito Chat supports text only, with image and voice support in development. Future tools like Side Chat will use the same Private Processing approach to bring contextual AI help into regular WhatsApp threads without exposing conversations to Meta.
What This Means for Privacy, Safety, and Trust
Incognito Chat marks a shift in how Meta handles user data in AI products: instead of logging conversations for model improvement or safety review, it discards them entirely. That appeals to people worried about private AI conversations being used for training, advertising, or future profiling. It also aligns with broader trends toward stronger WhatsApp encryption and privacy guarantees. However, full ephemerality raises hard questions. If there’s no stored record, harmful or dangerous conversations are harder to audit after the fact. Meta says safety guardrails still operate in real time—refusing harmful prompts, steering users toward helpful information, and even stopping interaction when necessary—but critics note the trade‑off: stronger confidentiality can create blind spots for moderation. Ultimately, Incognito Chat asks users to reconsider how much they trust Meta’s technical promises. Confidential computing reduces the need for blind trust, yet skepticism is likely to persist given past controversies around data use and AI training.