ChatGPT launches locked chats feature; resetting the PIN will permanently delete all locked conversations

On October 20, a lock appeared in the ChatGPT chat box.

OpenAI rolled out the "Locked Chats" feature: after setting a 4-digit account PIN, users can hide designated conversations.

On supported mobile devices, they can also access them via face scan, fingerprint, or the device unlock method.

Conversations you can't bear to delete but don't want outsiders to stumble upon now have a dedicated space.

But once a chat is hidden, several questions still need to be cleared up:

Will the AI still remember the secrets you've discussed? Will this content be used to train models? If you forget your PIN, will the chat history be wiped?

Locking a conversation does not mean all these questions are resolved.

In ChatGPT, hiding conversations, managing memory, and controlling training use each have their own rules.

So don't rush to stuff all your secrets in there—understand the boundaries of this lock before you use it.

Imagine this: you're screen-sharing in a meeting, and the sidebar shows your record of consulting about a health issue yesterday; you hand your phone to a friend to look at photos, but accidentally switch to a draft resignation letter; you want to prepare a holiday surprise for your family, but your chat list is full of gift lists and price-comparison records...

You still want to keep and continue using these private conversations, so you can't just delete them.

But when sharing your screen or lending out your device, a carelessly exposed chat snippet could make you "socially die" on the spot.

"Locked Chats" offers exactly this kind of compromise, and it's not complicated to use.

Follow the prompts to set and confirm a 4-digit PIN.

Locked conversations can be found in "Locked chats."

Once locked, the conversation is hidden from the regular history list, pinned section, and ordinary search results, and is instead moved into the "Locked chats" section under "Projects" in the sidebar.

After that, as long as you verify your identity, you can continue the conversation at any time.

According to the OpenAI Help Center, Locked Chats is gradually rolling out to the Free, Go, Plus, Pro, Business, Enterprise, and Edu plans, and free users can use it too.

Currently the feature is gradually covering the web, iOS, Android, and the new macOS and Windows desktop apps.

Of course, this lock is not all-powerful.

Temporary chats, archived chats, group chats, and conversations with attributes such as scheduled tasks or meeting recordings do not currently support locking.

Chats created via someone else's shared link, Codex, and local work conversations also have corresponding restrictions.

And the medical-specific ChatGPT for Healthcare simply does not offer the feature.

On the web, after clicking lock, the currently open tab may still continue to display that chat box; if you are also logged in and have the page open on a tablet or another computer, you must manually refresh to completely hide the title and loaded content.

Therefore, before you prepare to screen-share or hand your phone to someone else, never blindly trust the words "locked"—close the current page and refresh the list to truly complete the safeguard.

There is another detail that's easy to trip over: locking does not invalidate previously created share links, and anyone who has the link can still view it.

Content added during the lock period will not sync to the old link; if you want to revoke access, you need to separately delete this share link.

OpenAI also explains that locking only adds a viewing verification step; it does not provide end-to-end encryption, nor does it change data retention and model improvement settings.

It's more like a "screen privacy film": it reduces the chance of others seeing your chats, but the content is still handled under the original rules.

It can keep others out

Not so fast: putting "interface hiding" directly on par with "data isolation" is the biggest cognitive misconception.

First, look at the "memory" feature.

If you have memory enabled, locked chats may still draw on memories already saved in regular chats, so you don't have to reintroduce yourself.

Good news: new memories created while locked will not leak into regular chats. This means your contentious topics have a unique memory reserve of their own.

ChatGPT's "memory summary" page displays saved personal background information; after selecting content, you can click "Make a correction" to modify it.

But there is also an "old account" here:

For example, if a private matter was previously saved as a memory in a regular chat, then even after you lock that conversation, this memory may still be used in regular chats.

If one day you permanently unlock the conversation, the entire chat content may again be used to form regular memories according to account and workspace settings.

But memories formed separately during the lock period will not automatically merge into regular memories when unlocked.

In one sentence: old accounts are not written off just because you add a lock, while new accounts strictly follow "dedicated accounts for dedicated use."

Next, look at "training" data.

Locked chats do not change the original retention policy, nor do they automatically turn off model improvement; whether a conversation is used for training is still determined by the account's data control settings.

To stop new conversations from being used for model training, go to "Settings—Data controls" and manually turn off "Improve the model for everyone."

The model improvement toggle in Data controls.

Only after turning it off will new conversations not be used for model training.

But the official note adds: even if you turn off the model training toggle, if you actively submit feedback by liking, disliking, etc., the entire conversation related to that feedback may still be used to train models.

If you want conversations not to enter history, not to create or update memories, and not to be used for model training, you can use "Temporary Chat."

For security review purposes, conversations may still be retained for up to 30 days. If they are later saved as regular chats, they will be handled under regular chat rules.

Put simply, in ChatGPT, these functions each do their own job:

Forgetting your password is not scary

Randomly clicking reset is fatal

When using Locked Chats, there is one button with enormous destructive power: resetting the "PIN."

If you just want to change your old PIN, you can choose "Change PIN," and the chat history will be retained.

Even if you forget your PIN, there's no need to panic immediately: forgetting the password itself will not automatically delete chats or files.

What actually triggers deletion is choosing "Reset PIN" and confirming execution.

According to OpenAI's rules, resetting will permanently delete all locked chats in the account (not just the current one), memories saved only in locked chats, and files protected only by the old PIN.

This is an irreversible clearance, and even OpenAI customer service cannot restore deleted content.

Therefore, before pressing confirm, read the deletion scope carefully and confirm whether there are important conversations and files you need to keep.

You use it to write weekly reports and check code, and you also discuss health, relationships, and those difficult decisions with it.

Some questions are over once asked, but some conversations leave important thoughts worth revisiting again and again.

Locked Chats lets this content be kept while reducing the chance of others seeing it.

But we also need to know: once a chat is locked, are old memories still there? Can shared links still be opened? When you reset the PIN, which conversations and files will be deleted?

These consequences should be explained clearly before operation, not left for users to find answers in the Help Center only after they lose content.

References:

This article is from the WeChat public account "Xinzhiyuan" (ID: AI_era), author: ASI Revelation, editor: Yuanyu, published by 36Kr with authorization.