ChatGPT adds locked chats that hide conversations behind a 4-digit PIN

What OpenAI launched

OpenAI has started rolling out locked chats, a ChatGPT setting that moves chosen conversations out of the normal sidebar and puts them behind a 4-digit PIN. The feature appeared in ChatGPT's release notes on 20 October 2026 and comes with a long help center article that is unusually frank about what the lock does and does not protect.

Once a chat is locked, its title and content disappear from regular chat history, pinned chats and conversation search. The conversations collect in a project called Locked chats, which opens only after the user enters the account PIN, or uses Face ID, Touch ID or the phone's own unlock method on supported mobile devices.

OpenAI pitches it for everyday privacy rather than security against attackers: keeping a sensitive conversation out of view when someone shares their screen, hands their phone to a friend or uses a shared computer. TechCrunch, which reported the launch, notes OpenAI's examples of discussing something personal or planning a surprise.

The rollout is gradual. The help article lists the feature for ChatGPT Free, Go, Plus, Pro, Business, Enterprise and Edu, while the release note mentions Free, Go, Plus and Pro on the web, the iOS and Android apps and the new ChatGPT desktop app for macOS and Windows. Users who do not yet see a Lock chat option in a conversation's menu may have to wait a few days.

How locking a chat works

Locking starts from the three-dot menu of an existing conversation. ChatGPT asks the user to create a 4-digit PIN the first time, then moves the chat into Locked chats. New conversations can then be started directly inside that area, and they never touch the regular history.

The PIN belongs to the OpenAI account rather than a device or a workspace, so the same code works on every device signed in to that account. Face ID, Touch ID or Android device unlock is optional and has to be turned on separately on each phone. After too many wrong guesses, ChatGPT temporarily blocks further attempts.

Access is deliberately short-lived. On iOS and Android it ends after one minute away from the locked area or one minute with the app in the background. On the web it is tied to a single browser tab and ends when the tab is refreshed or closed; on the desktop app it is tied to one window. A Lock now button ends access immediately. On Android, screenshots of locked chats come out blank because the app hides protected content.

There are rough edges the help page spells out. Other devices that already have ChatGPT open may keep showing a newly locked chat's title until their lists refresh, and on iOS previously loaded content can stay visible until the app reloads. It can also take time for search results to update. OpenAI's advice is to check other open devices before sharing a screen.

What a lock does not do

The most useful part of the documentation is a list of things a lock does not change. Locking does not create end-to-end encryption or a separate encrypted vault. It does not change how long OpenAI keeps the conversation, and it does not change the Improve the model for everyone setting, so a locked chat is used for training exactly as it would have been before.

It also does not erase anything ChatGPT already remembered from that conversation before it was locked, and it does not override workspace policies, safety requirements or legal and compliance obligations. Engadget summed up the point plainly: locked chats are not a way to get around OpenAI's legal obligations. Anyone who can open a locked chat can still copy what is in it, and OpenAI notes that ads, where they are shown to an account, can also appear in locked chats.

In other words, the lock is a screen-level privacy control. It protects against the person looking over a shoulder, not against a subpoena, a breach or a company administrator. That distinction matters more as ChatGPT itself becomes an advertising surface, a shift AI Lookout covered when OpenAI opened custom audience targeting to advertisers.

The PIN reset deletes everything

The design choice most users will need to understand is what happens to a forgotten PIN. Forgetting it does not delete anything. Resetting it does: a reset permanently deletes all locked chats, any memories stored only inside locked chats, and files protected only by the old PIN. The help page says the deletion cannot be undone and that OpenAI Support cannot recover the content.

A workspace administrator cannot reset a member's PIN or unlock their chats either. Changing the PIN, as opposed to resetting it, keeps everything, but it requires entering the current PIN even for people who normally use Face ID or Touch ID. After a reset, the deleted conversations follow OpenAI's standard retention policy for deleted chats.

That makes the PIN closer to a key than a password with an account-recovery fallback. It is a sensible trade if the goal is that nobody, including support staff, can hand the locked conversations to someone who simply gains control of the account. But it means a user who locks months of important work and then forgets four digits loses it.

Memory, files and shared links

Locked chats get their own memory. When Memory is turned on, locked conversations can still use what ChatGPT knows from regular Memory, and new information from messages sent after locking can be remembered for use in other locked chats. That information does not flow back into regular chats, and memories created inside the locked area are not added to regular Memory if the chat is later unlocked.

Files uploaded or created in a chat are protected when it is locked, including files added afterwards. They become visible only inside that chat and are hidden from Space, ChatGPT's file area, and its search. Files that were brought in from elsewhere in Space stay available there, and files from connected sources keep their existing access rules.

Shared links need care. A chat that already has a share link can still be locked, but locking does not delete the link: the previously shared content stays public, although new messages are not added to it. Users have to delete the link separately. The point is timely. On the same day, Meta paused sharing in its Muse app after researchers showed that Muse share links could be found through public DNS records, exposing thousands of conversations.

Locking a chat that lives in a project also pulls it out of that project, so it stops using the project's instructions, files and memory. A permanently unlocked chat returns to regular history rather than to its old project.

Work accounts and compliance

For companies, the lock is not a way for employees to hide conversations from the employer. Locked chats do not override an organisation's access to workspace data. In ChatGPT Enterprise and Edu, locked chats are still included in the conversation message logs available through the OpenAI Compliance Platform.

Administrators also cannot switch the feature off: the help page says there is no workspace setting or role-based permission to disable locked chats. On the other hand, they cannot unlock an employee's chats or reset the employee's PIN from the admin console, which keeps the lock meaningful against casual access by colleagues while leaving formal compliance access intact.

Which chats cannot be locked

Several kinds of conversation cannot be locked at all: temporary chats, archived chats, group chats, chats in the earlier Health experience, chats created from a shared link, chats with scheduled tasks or meeting recordings, chats tied to background tasks and chats whose responses are saved as project sources. ChatGPT for Healthcare does not get the feature.

On the desktop app, cloud chats in Chat or Work mode can be locked, but Codex conversations, including cloud Codex, and local Work conversations cannot. Voice works inside locked chats on the web and in the mobile apps but not in the desktop app. Work mode has grown quickly this autumn, including a Chrome extension that reads and controls open tabs, so the exclusion of local Work sessions is a notable gap for people who use ChatGPT on their own files.

OpenAI also draws a line between locked chats and temporary chats. Temporary chats stay out of history and are not used to improve models while they remain temporary, but they are not saved for later. Locked chats are saved and protected by the PIN, and they follow the user's normal model-improvement setting.

Why it matters and what is missing

Phone makers have offered locked folders for photos and messaging apps for years, and as people use chatbots for health questions, money and relationships, a chat history has become just as sensitive. OpenAI's version is careful to promise only what it delivers: a hidden, PIN-gated place in the interface.

What it does not offer is the stronger protection some users may assume. There is no end-to-end encryption, no separate retention rule and no automatic exclusion from training. Users who want a conversation kept out of model training still have to change that setting or use a temporary chat, and users worried about stored data still face the same retention policy as before.

Some details are still unclear. OpenAI has not said how long the temporary lockout after wrong PIN attempts lasts, and the release note and help page differ on which business plans have it today. AI Lookout has not tested the feature, so the behaviour described here comes from OpenAI's documentation and the launch reports.