In a fresh build of the ChatGPT Android app, version 1.2.26.3, researchers discovered code fragments describing a new tool called Locked Chats. This is reported by RBC-Ukraine, citing Android Authority. The feature is designed to protect a user's private conversations with the neural network: locked dialogs will open only after passing authentication — via a personal PIN code or a biometric fingerprint scan. In doing so, OpenAI is effectively bringing into its chatbot the level of privacy that is familiar from messengers and banking apps.
How exactly the protection will work
According to the descriptions found in the code, unlocked conversations completely disappear from the side panel, from the general history, and from the app's internal search results. In other words, neither a casual glance at the screen nor a search by keywords will allow a stranger to discover a "locked" dialog. To access it, the user will have to authenticate each time. Notably, even after a successful unlock, the chat does not automatically become public: it remains in a protected state until the user manually removes this status. This eliminates the risk that a dialog left open accidentally ends up in the general feed.
Difference from Temporary Chats mode
ChatGPT already has a Temporary Chats mode, however it has a significant drawback: such dialogs are not saved at all, and it is impossible to return to them later. The new Locked Chats feature fills exactly this gap — the conversation remains in the account and is available for further use, but at the same time it becomes hidden from prying eyes. In essence, this is a compromise between full deletion of history and its full openness: the data is stored, but isolated from general access.
Isolation of the neural network's memory
The developers paid special attention to how the memory algorithms work. Protected chats may use previously formed general account memory, that is, rely on what the AI already knows about the user. At the same time, all new information that the model learns during a "private" dialog will be isolated in a separate protected environment. This data will personalize only other locked conversations and will not end up in the regular history. This approach allows preserving the privacy of sensitive topics without depriving the neural network of context.
Contradictory data
It is important here to honestly distinguish two viewpoints. On the one hand, a number of sources (including Android Authority and RBC-Ukraine) present Locked Chats as a capability discovered in the code, that is, a feature that has not yet been officially released and is not confirmed by an OpenAI release announcement. On the other hand, some publications phrase the news as "hidden chats will appear in ChatGPT," which reads like a finished product. The difference is significant: at the time of preparing this material, the discussion concerns code fragments in build 1.2.26.3, not a tool available to all users. The developer has not announced the dates or conditions of the official launch, so the exact timing of the feature's appearance remains unknown.
Context: expansion of the ChatGPT ecosystem
The launch of Locked Chats fits into OpenAI's broader strategy of expanding ChatGPT's capabilities across different platforms. Recently, the company presented a new plugin for the Mac version of ChatGPT, allowing the AI to control the Apple Messages app. Extensions are used through the system mechanisms of Codex and ChatGPT Work, and for the plugin's operation the developers employed current macOS system tools — AppleScript and accessibility elements (Accessibility). The totality of these steps indicates that OpenAI is steadily transforming ChatGPT from a simple chat interface into a full-fledged application with its own privacy mechanisms and integration with the operating system.
