A single conversation inside ChatGPT can now be walled off from whoever picks up the phone next. OpenAI began rolling out Locked Chats on Tuesday, sealing a thread behind a four-digit PIN or a look from Face ID or Touch ID on hardware that supports it.
Locked conversations shift into their own section. They drop out of ordinary history, out of pinned items and out of search results, and the user can reopen them whenever they like. That separates them from Temporary Chats, which cannot be revisited. The pitch is the borrowed-phone moment: someone shows a relative a picture in one thread while a medical discussion or a planned surprise sits a couple of taps away.
Memory behaves differently here too. The confined section keeps its own store, shared only among conversations locked behind it, so details written there never surface in the main memory that ordinary chats draw on.
What the lock does not do
OpenAI is blunt about the limits. No end-to-end encryption comes with it, and no separate encrypted vault. Retention rules are unchanged, whatever the model remembered before the lock went on stays remembered, and the broader “improve the model for everyone” switch is untouched.
A plan for the worst case
Changing the PIN is no soft reset. Reset the code and every chat, file and item it protected is destroyed permanently, a firm anti-guessing default and a trap for anyone who rotates digits casually.
Subscribers on every paid tier get it, plus Free, Go and Edu accounts, though ChatGPT for Healthcare is excluded outright. Neither are temporary, archived or group threads, ones opened from a shared link, or any carrying scheduled tasks or meeting recordings. Because rollout started Tuesday, some accounts will not see the option yet.
