Four-digit PINs now gatekeep private AI chats
OpenAI is rolling out a feature that hides selected conversations from standard search and history, though it does not add end-to-end encryption.
By Andrew Kovalenko ·
Users who share a device with a partner, family member, or colleague can now hide specific ChatGPT conversations from casual view. OpenAI has introduced Locked Chats, a tool that moves selected dialogues into a separate folder secured by a four-digit PIN. The feature aims to prevent accidental exposure of personal advice or planning details when a phone or computer is handed to someone else.
The rollout is gradual and covers Free, Go, Plus, Pro, Business, Enterprise, and Edu accounts. It functions on web browsers, the ChatGPT desktop apps for Mac and Windows, and mobile applications for iPhone and Android. OpenAI stated that the dedicated ChatGPT for Healthcare platform does not support this feature. Users on supported mobile devices can opt to use Face ID, Touch ID, or the phone's native unlock method instead of typing the PIN every time.
To lock a conversation, a user opens the three-dot menu on any chat and selects the lock option. The system creates a single PIN for the entire account, which works across all signed-in devices. Once locked, the conversation vanishes from the main sidebar, pinned chats, and standard search results. It resides under the Projects section in a specific Locked chats folder. Users can continue the conversation from this protected space without moving it back into the general history.
The feature has specific implications for existing workflows. If a user locks a conversation that is part of a ChatGPT Project, the chat is removed from that project. It loses access to the project's specific instructions, files, and memory. If the user unlocks the chat later, it returns to the general history, not the original project. This disconnect may require users to reattach context or files if they rely on project-specific data.
OpenAI clarified that Locked Chats are not an encrypted vault. The company does not use end-to-end encryption for these folders. Locking a chat also does not alter data retention policies or the user's existing settings for model improvement. If a user has previously shared a conversation via a link, that link remains valid after the chat is locked. Users must manually delete the shared link to stop others from accessing the content. Files uploaded within a locked chat receive the same protection, but files already stored in ChatGPT Space or connected services keep their original access permissions.
Automatic locking helps mitigate the risk of a lingering screen. On mobile devices, access to a locked chat ends after one minute of inactivity or if the app moves to the background. On the web, closing the tab or refreshing the page ends access. Users can also select a Lock now button to secure the session immediately. A temporary glitch may leave the conversation title visible in the browser tab or on other devices until they refresh.
For workplace accounts, organizational compliance policies still apply. Enterprise and Edu administrators cannot unlock chats or reset PINs, but authorized compliance systems may still access conversation logs. This means that for highly sensitive data, such as financial credentials or personal identifiers, the PIN provides a layer of convenience rather than a complete security barrier.
Users concerned about AI training can separately toggle off the Improve the model for everyone setting in their data controls. This prevents new conversations from being used to train OpenAI's models. Business, Enterprise, and Edu workspace content is not used for model training by default, according to the company.
