OpenAI shipped ChatGPT locked chats on 20 October 2026. You put a four-digit PIN on a conversation and it leaves your chat history, your pinned list and your search results. It does not leave OpenAI, and that gap is the whole story for anyone who pastes client work into the box.
You lock from the chat’s more options menu, create or enter a four-digit account PIN, and the chat moves under Projects into a Locked chats section. Face ID, Touch ID or Android device unlock can stand in for the PIN, set up separately per device.
The chat then stays out of regular history, pinned chats and ordinary search. You cannot search across locked chats at all, so finding one means browsing that section by hand. Access is short-lived by design: on the web it ends when you refresh or close the tab, and on iOS and Android after one minute away or in the background. None of that closes the obvious hole, which OpenAI names itself. Anyone who can open a locked chat can still copy what is inside it.
The release notes of 20 October say the feature is rolling out gradually to Free, Go, Plus and Pro, on the web, on iOS and Android, and in the new desktop app for macOS and Windows. The help page for the feature lists Free, Go, Plus, Pro, Business, Enterprise and Edu, and excludes ChatGPT for Healthcare. The plans most likely to hold client data were documented but not announced, and that matters because of what the same page says about them further down.
The release note puts it in one line: your retention and model-improvement settings stay unchanged. The help page is more specific. Locking does not create end-to-end encryption or a separate encrypted vault. It does not change how long OpenAI keeps the chat. It does not remove anything ChatGPT remembered before you locked it. It does not change your training setting.
There is a fifth limit, and on a team plan it is the one that counts. Locking does not override workspace policies, safety requirements, or legal and compliance obligations. On Enterprise and Edu, locked chats remain in the conversation message logs available through the OpenAI Compliance Platform, and an administrator cannot disable the feature through a workspace setting or a role permission. Your admin cannot unlock your chat, and also does not need to. This is a screen-privacy feature, built for the moment you hand somebody your laptop.
OpenAI’s data controls page says that on personal Free, Plus and Pro workspaces, data sharing is enabled by default. The setting is called “Improve the model for everyone”, and if you have never opened it, it is on. A locked chat inherits it, so the PIN stops the person beside you reading the conversation while the conversation keeps doing what it already did. That is the same shape of trap as the privacy condition attached to Google’s saved instructions in September, where a useful feature carried a data requirement almost nobody read.
Turning the setting off does not tidy up history either. The same page says it does not delete or hide saved chats, only stops new conversations being used for training. Check whichever account you actually draft in, including a free plan you keep for overflow.
Temporary chats are the tool for the job the PIN looks like it does. They stay out of history, do not create or update memories, are not used to improve models, and the page says they “may be retained for up to 30 days for safety purposes”. That is the only retention figure OpenAI publishes across any of these pages. A locked chat is one you keep and hide. A temporary chat is one you do not keep.
Forgetting the PIN costs nothing on its own. Resetting it costs everything behind it. The help page is blunt: resetting your account PIN permanently deletes all locked chats, along with memories stored only within them and any files protected only by the old PIN. It adds that this cannot be undone and that OpenAI Support cannot recover the deleted content.
There is no administrator route around it. Workspace administrators cannot reset your PIN or unlock your chats, which is good for privacy and bad on the day somebody leaves the agency with three months of campaign reasoning behind four digits nobody else has. Changing the PIN is safe and keeps the chats, but it needs the current PIN even if you normally unlock with a fingerprint. So the housekeeping on day one is recording the PIN somewhere that is not ChatGPT.
Locking a chat that sits in a project removes it from that project, so it stops drawing on the project’s instructions, files and memory. Unlocking returns it to regular history, not to the project it came from. If a project holds your brand voice rules or a client’s style guide, locking a chat inside it quietly disconnects the part of your content workflows that made the output usable.
Several ordinary actions are blocked while a chat is locked. You cannot archive, pin, move or share it without unlocking first. Scheduled tasks have to be deleted before you lock, paused ones included, and a locked chat cannot create or update them. An existing shared link survives locking: content already shared stays reachable and new messages simply stop being added, so revoking access still means deleting the link yourself.
Memory runs as a second, separate store. Anything learned after locking is remembered only for other locked chats, does not carry into regular chats, and is not added to regular Memory when you unlock. One last detail says plenty about what this feature is for: ads can appear in locked chats if ads are available to your account.
No. OpenAI states that locking does not create end-to-end encryption or a separate encrypted vault. It is an access control on the conversation list, not a change to how the data is stored. TechCrunch reported the launch on 20 October and made the same point.
No. Locking leaves the “Improve the model for everyone” setting exactly as it was, and on personal Free, Plus and Pro accounts that setting is on by default. To keep a conversation out of training, change the setting or start a temporary chat.
Nothing, until you reset it. A reset permanently deletes every locked chat plus any memories and files that existed only behind that PIN, and OpenAI Support cannot recover them. Administrators cannot reset it for you either.
On Enterprise and Edu the conversation stays in the message logs available through the OpenAI Compliance Platform, and administrators cannot switch the feature off. They cannot open your locked chat in the app, but the organisation’s logging obligations are unaffected by it.
If you demo ChatGPT on a shared screen or run client and personal work in one account, the PIN is worth setting up this week. Pair it with the training setting, which is the control that decides what happens to the conversation after you close the tab. Then work out what the plan you are on is allowed to hold, and keep the genuinely sensitive material somewhere built to hold it.
This is the kind of thing we work on with clients at Social Lady. If you want a second opinion on what your team should and should not paste into an AI tool, get in touch.
