Meta introduces WhatsApp incognito mode for AI chats

Meta Platforms rolled out an "incognito" mode for WhatsApp so users can have private conversations with the app's AI assistant, the Associated Press reported on Wednesday. WhatsApp's official product page and help documentation describe Incognito chats as private, encrypted conversations that are processed in a secure environment and disappear when the chat is closed, per WhatsApp's public pages. WhatsApp's FAQ explains that the feature is built on Private Processing, which the company documents as running on a Trusted Execution Environment in the cloud and not retaining messages after processing.

WhatsApp's help center and FAQ state that Private Processing routes only the data needed to fulfil a request to the secure environment, encrypts requests between device and processing, and does not retain the messages or responses after completion. The FAQ adds that users can create and download a report of Private Processing activity on their device, and that some AI features using Private Processing are optional and available in a limited set of countries and languages. Meta's help pages also document deletion flows for Meta AI content and include a note that interactions with AIs may be used to improve Meta's models; the help documentation states that when AIs cannot answer a request, Meta may share messages and general information, like region, with select partners to improve results.

Messaging platforms and AI vendors have increasingly added privacy-preserving modes and ephemeral options as a product response to user concerns about sending sensitive text and media to cloud AI services. Observed patterns in similar product designs emphasize a tradeoff: stronger cryptographic and enclave protections limit telemetry available for model improvement while preserving user confidentiality for sensitive flows.

For practitioners: Engineers building or integrating conversational AI in private messaging should expect to contend with integration complexity around Trusted Execution Environment APIs, encrypted transport, and ephemeral storage guarantees. Companies implementing comparable features typically need robust logging policies, secure key management, reproducible evaluation pipelines that do not rely on long-term retention of user prompts, and clear user-facing controls for opting in or out of model-improvement data collection.

WhatsApp's documentation notes feature availability limits, currently supported languages include English, Indonesian, Portuguese, and Spanish, and that Private Processing features may not be available everywhere. WhatsApp's help material also cautions that an AI's responses may be inaccurate or inappropriate and explains the optional nature of Private Processing features. The help pages further state that Meta may use interactions with AIs to improve its systems and, in some cases, share messages and metadata with select partners when AIs cannot answer a query.

Observers will monitor:

• •whether security audits or third-party assessments confirm the enclave and Private Processing guarantees
• •how retention and telemetry policies are implemented in practice and surfaced to end users
• •whether competing platforms introduce similar ephemeral-AI modes or stricter data-use controls. For practitioners: watch developer documentation for APIs, keys, and export controls, and monitor regulatory guidance in jurisdictions that treat algorithmic training data and private communications as sensitive information