ChatGPT Locked Chats Hide a Conversation. They Do Not Turn It into a Vault

A lock icon can imply more than a feature actually provides. OpenAI’s new Locked Chats option adds a useful layer between a saved conversation and someone casually browsing an already signed-in account. But it does not convert ChatGPT into a separate encrypted vault. The details matter most when a conversation contains something sensitive enough that a user expects the lock to change who can access it, how long it is kept or whether it can influence personalization.

What the lock changes

OpenAI says locked chats move selected eligible conversations out of regular history, pinned chats, search and previews. Users open them from Projects > Locked chats with an account PIN; supported mobile devices can also use Face ID, Touch ID or device unlock. The feature is rolling out gradually, and available conversation types vary by platform. ^[1]

That can be useful in ordinary situations: keeping a personal planning thread out of a shared screen, separating a saved conversation from the usual sidebar or reducing accidental discovery on a borrowed device. The control is about visibility and access within the account experience. It is not a statement that the underlying conversation has been encrypted separately or removed from the service’s ordinary data handling.

OpenAI lists several things locking does not do. It does not create end-to-end encryption, change retention, change the “Improve the model for everyone” setting, remove information the system remembered before locking, or override workspace rules and legal obligations. Enterprise and Edu chats remain included in message logs available through the OpenAI Compliance Platform. ^[1]

Keep privacy controls conceptually separate

People often bundle several expectations under the word “private.” At least four distinct controls matter here: who can see the conversation in the interface, whether the provider retains it, whether eligible data can be used to improve models and whether the system can use information to personalize later responses. A chat lock addresses the first and adds an authentication step for access. It does not automatically change the other three.

OpenAI’s data-controls documentation says the model-improvement setting is separate from chat history. Turning off “Improve the model for everyone” affects whether new personal-workspace conversations are used for training, but does not delete or hide saved chats. Memory is also separate from model-training controls. Managed workspaces can have their own retention, memory, compliance and access policies. ^[2]

This distinction helps users choose the right tool. Use a locked chat when the goal is to keep a saved conversation out of ordinary navigation and require authentication to reopen it. Use data controls to manage model-improvement preferences. Review memory settings if the concern is what may carry into future conversations. For regulated or employer-managed information, follow the organization’s approved workspace and retention policy rather than assuming a personal interface control changes it.

The PIN has a consequential recovery rule

The account PIN is not just a convenience. OpenAI says resetting a forgotten PIN permanently deletes locked chats, memories stored only within those chats and files protected only by the old PIN. Support cannot recover that deleted content. Changing the PIN while you still know the current one does not delete the chats; resetting is different from changing. ^[1]

Before locking important material, decide whether the conversation should be stored in ChatGPT at all and keep a separate export or approved record if the content must remain available. Do not treat the PIN as a recovery mechanism. If a team needs durable business records, use its documented retention and export process, and confirm who owns that process.

There are also device-level details. OpenAI warns that other devices with ChatGPT already open may temporarily continue to show a title until lists refresh; on iOS, previously loaded content may remain visible until the app refreshes. On web, the locked chat can remain open in the current tab after locking. Users should refresh or close other sessions before sharing a screen or handing over a device. ^[1]

A safer way to describe the feature

“Hide and lock” is a more accurate mental model than “secure vault.” The first describes what the product says it does. The second can imply encryption, retention isolation or privileged access boundaries that OpenAI explicitly says this feature does not provide.

For a personal account, check the model-improvement setting and memory behavior separately, then lock eligible conversations when access visibility is the concern. For an Enterprise or Edu account, ask the administrator how compliance logs, retention and legal holds apply. For any high-consequence information, avoid placing it into a consumer assistant unless the relevant policy and contract permit it.

The feature is still rolling out, so availability and behavior can vary. OpenAI’s help documentation is the best current authority for supported platforms and exceptions; recheck it before relying on the flow. A lock is a useful interface safeguard when described precisely. It becomes misleading only when people assume it changes the data policies around the conversation.

Read next

• Enterprise ai privacy data custody safeguards
• Local ai agents active parameters privacy