ChatGPT just got a lock, and the fine print matters more than the feature. OpenAI announced Locked Chats on Tuesday, letting users protect individual conversations with a four-digit PIN, Face ID, or Touch ID. It’s a screen-privacy tool. It is not a security upgrade.
Here’s how it works. Lock a chat and it moves into a separate Locked Chats section, away from regular history, pinned conversations, and search results. Unlike Temporary Chats, which vanish into a separate session, you can come back to locked chats whenever you want. OpenAI’s pitch is modest: you’re planning a surprise, or discussing something personal, and then you hand your phone to a friend to show them something.
The memory handling is the smartest part. With Memory on, Locked Chats gets its own memory pool, shared only across locked conversations. New messages in locked chats don’t feed regular Memory, so a sensitive topic won’t show up in a casual chat later. Anyone who has seen ChatGPT surface an old detail at an awkward moment will appreciate that.
Now the limits. OpenAI says locking a chat does not create end-to-end encryption or an encrypted vault. It doesn’t change how long the chat is retained. It doesn’t erase anything ChatGPT remembered before the lock. And it doesn’t touch the “Improve the model for everyone” setting, so a locked chat can still be used for training if that’s switched on. Resetting your PIN permanently deletes all locked chats and files tied to the old one. Several chat types are excluded too: temporary, archived, group, shared-link, scheduled-task, and meeting-recording chats.
Why does this matter? People now use ChatGPT for health worries, money problems, relationship advice, and job searches. That’s diary territory, on a device that gets passed around. Apple and Google already let you lock notes and hidden photo folders, so a lock on chat history is overdue. Rivals like Claude and Gemini haven’t made much of this kind of local privacy control, which gives OpenAI a small edge on consumer trust.
But don’t mistake a lock for privacy in the deeper sense. The real questions are what OpenAI stores, for how long, and whether it trains on it. None of those answers changed. If you need that, use Temporary Chats, turn off training, or run a local model.
The rollout starts Tuesday across Free, Go, Plus, Pro, Business, Enterprise, and Edu plans, but not ChatGPT for Healthcare. Useful, overdue, and narrower than the name suggests.
