OpenAI has begun the phased rollout of Locked chats, a new privacy feature in ChatGPT that allows users to hide selected conversations from their general chat history. Protected dialogues are isolated from the sidebar, the list of pinned chats, and standard in-service search. Access to them is granted via a four-digit account PIN on desktop platforms or biometric authentication—such as Face ID, Touch ID, or standard device unlocking—on iOS and Android mobile devices. The feature is positioned as a solution for device-sharing scenarios, screen-sharing with colleagues, or working in public spaces where users do not want outsiders seeing the contents of their interactions with the AI.
How the Protection Works and Where to Find Hidden Chats
Protected conversations do not vanish from the system permanently: they are gathered in a dedicated folder under the "Projects — Protected chats" section. It is from there that users can return to a conversation by entering their PIN or completing biometric verification. On the desktop versions for macOS and Windows, authentication is performed exclusively via a four-digit code, while the mobile apps for iOS and Android support the full suite of biometric methods available on the specific device. This approach preserves privacy without requiring users to memorize separate passwords for each individual dialogue.
Availability: All Subscription Tiers and New Desktop Apps
The Locked chats feature is rolling out to users across all active subscription tiers—Free, Go, Plus, Pro, Business, Enterprise, and Edu. In addition to the web interface and mobile apps, OpenAI has integrated chat protection into the new desktop apps for macOS and Windows, making the tool available across all major platforms where ChatGPT operates. The phased rollout means the feature may appear for users on different days, and not all accounts receive access simultaneously.
Technical Nuances: Files, AI Memory, and Limitations
Dialogue protection affects not only the chat text but also accompanying data. Documents uploaded to a protected chat are hidden from general file storage and remain accessible exclusively within that conversation. Regarding the AI's memory, a protected chat can utilize previously saved information from the user's general profile; however, new facts captured during the locked session are not transferred to the general memory. This means the context built within a hidden dialogue will not influence subsequent open conversations. Functional limitations include: closed chats cannot be archived, pinned in the sidebar, or shared via links. The feature also does not apply to temporary and group chats.
Important Caveats: Lack of End-to-End Encryption and Risk of Permanent Data Loss
OpenAI explicitly states that the implementation of a PIN does not introduce end-to-end encryption and does not alter the company's overall data retention policy. This means that server-side information is still stored in accordance with standard procedures, and the PIN acts as client-side protection against unauthorized viewing on a specific device. A critically important detail is that if a user forgets their PIN, resetting security settings will lead to the permanent deletion of all hidden chats, associated files, and local memories. Neither OpenAI support nor enterprise workspace administrators will be able to recover this data. The company recommends that users carefully consider which dialogues to place in protected mode, given the impossibility of recovering them after a reset.
Context: Privacy as a Competitive Advantage
The launch of Locked chats continues OpenAI's trend of expanding data control tools, which is especially relevant amid growing competition from Anthropic, Google, and other generative AI developers. In the corporate segment, where ChatGPT Enterprise and Business tiers are actively integrated into workflows, the ability to isolate sensitive conversations from colleagues using the same device becomes not just a convenience, but an information security requirement. At the same time, OpenAI maintains transparency in describing the feature's limitations, setting the company's approach apart from the marketing promises of "absolute privacy" frequently found in competitor press releases.
