Meta has rolled out an incognito chat mode for its artificial intelligence assistant on WhatsApp, processing messages in a secure environment that remains invisible even to the company itself. The feature, announced this week, directly confronts longstanding unease over the data-hungry nature of generative AI systems, as users increasingly share financial, medical, and personal details with chatbots. Under the new setting, conversations are not saved by default and disappear as soon as the session ends, offering what Meta describes as a confidential space for exploration without surveillance.

Viewed from Washington, the move reflects a belated recognition that large language models, trained on vast troves of data, have eroded user trust. Will Cathcart, WhatsApp’s head, noted during a press briefing that people frequently entrust AI with information they would never want stored indefinitely. From Latin America, where the feature has already been activated in Argentina, early responses suggest a pragmatic embrace of ephemeral AI chats, though some users remain sceptical about what “invisible” processing truly entails. The company insists that the private processing technology prevents even Meta from accessing the exchanged data, a claim that will likely face scrutiny from privacy advocates.

In Europe, the announcement arrives against a backdrop of tighter regulatory pressure. Under the Digital Markets Act, Meta has already introduced interoperability for third-party messaging apps on WhatsApp, as analysts in Milan have detailed. While the incognito mode is not mandated by the DMA, it aligns with a regional push for greater user control over data. Observers in Brussels interpret the feature as a defensive manoeuvre to pre-empt stricter privacy rules, particularly as the European Union’s AI Act begins to take effect. The convergence of regulatory demands and consumer expectations is forcing tech giants to rethink how they balance data collection with consent.

The incognito chat is part of a broader industry recalibration. OpenAI, for instance, recently unveiled a “trusted contacts” feature for ChatGPT that alerts designated individuals if a user appears to be in mental health distress during a conversation. This parallel development underscores a wider effort to embed safety and privacy safeguards directly into AI interactions, addressing both ethical concerns and potential litigation. For Meta, the new mode may become a template for its other platforms, including Messenger and Instagram. Yet questions remain about how such features will be independently audited and whether ephemeral processing can truly prevent data leakage. The next challenge for regulators and users alike will be verifying that these private spaces are as secure as promised, and that incognito does not become a veil for opacity.