OpenAI has begun rolling out locked chats in ChatGPT, a feature that hides individual conversations behind a four-digit account PIN so they disappear from the main history, from pinned chats and from ordinary search. The release notes are dated October 20, and the accompanying help-centre article is strikingly candid about what the lock does — and, more importantly, what it does not do.
- A visibility layer, not a vault: locked chats are regrouped into a Projects area called Locked chats. No end-to-end encryption is added, and nothing moves into a separate encrypted container.
- Almost every plan: Free, Go, Plus, Pro, Business, Enterprise and Edu. ChatGPT for Healthcare is the single explicit exclusion.
- Reset means deletion: a forgotten PIN can only be reset, and the reset permanently wipes every locked chat, the files protected by it and any memory stored only inside it.
- Workplace logs stay intact: in Enterprise and Edu workspaces, locked conversations still appear in the message logs of OpenAI’s Compliance Platform.
How locking works across devices
The mechanism is simple. Open the three-dot menu on any eligible conversation, choose Lock chat, and ChatGPT asks you to create a PIN the first time. That PIN belongs to your OpenAI account rather than to a single device or workspace, so it travels with you across the web, the iOS and Android apps, and the new desktop application for macOS and Windows. The rollout is gradual, which means the option may not have reached your account yet.
Biometrics are a convenience layer, not a replacement. Face ID and Touch ID on iPhone, or the standard device unlock on Android, can open the locked area, but each device must be configured separately — and changing the PIN still requires typing the current one. Too many failed attempts trigger a temporary lockout.
One detail worth noting for organised users: locking a conversation that lives inside a project pulls it out of that project, and it stops using that project’s instructions, files and memory. Voice input works on the web and in the mobile apps, but not in the desktop client, and Android may render screenshots of a locked chat as blank or black.
What the lock does not protect against
OpenAI’s own documentation lists the caveats rather than leaving them to inference. Locking does not create end-to-end encryption or a separate encrypted vault, does not alter how long the conversation is retained, and does not change the Improve the model for everyone setting. If that toggle is on, a locked chat can still be used to train the company’s models. Anything ChatGPT memorised before you locked the chat stays in Memory, and workspace policies, safety requirements and legal obligations continue to apply.
Advertising follows the same logic. If your account sees ads, they can appear inside a locked conversation, and locking changes no advertising preference. The clearer illustration of the gap is the audit trail: what you hide from the person beside you remains fully visible to compliance reviews.
The forgotten-PIN trap
Changing a PIN is safe, provided you still know the old one. Forgetting it deletes nothing on its own — but resetting it is irreversible.
“OpenAI Support cannot recover the deleted content.”
That single sentence carries more weight than the marketing framing suggests. Workspace administrators cannot reset a colleague’s PIN, unlock someone else’s chats, or switch the feature off from a workspace setting.
Access also expires quickly: on the web and desktop, leaving the tab, refreshing or closing the window re-locks everything; on mobile it ends roughly a minute after you leave the locked content or background the app. A Lock now button cuts access immediately.
Files are confined to the conversation they were created in. They do not appear in Space or in any other chat — including other locked chats — even with the PIN entered. Shared links are the easiest leak to overlook: locking a chat that already has a public link does not delete that link, and previously shared content stays reachable, although new messages are not added while the chat is locked. The link must be removed separately, and a chat created from a shared link cannot be locked at all.
The exclusion list is long: temporary chats, archived conversations, group chats, the legacy Health experience, chats with scheduled tasks, meeting recordings or background tasks, plus Codex conversations and local Work chats. Locked conversations can no longer be archived, pinned, moved, shared or searched within the locked set — only browsed.
Read against the feature’s real purpose, the design is defensible and the framing is not. Handing a phone to a colleague to show a photo has always meant exposing whatever sat one tap away, and a PIN genuinely fixes that. But a screen-level barrier is a much smaller promise than protecting sensitive material, and OpenAI’s documentation is more honest than much of the coverage: content is retained on the same terms, training preferences are untouched, and ads can still surface. Shipping an access PIN well before any real encryption option is a revealing order of priorities. For anyone who uses ChatGPT for genuinely delicate work, the lock protects the audience in the room — not the data on the server.