What Locked Chats Tell Us About Privacy UX

People now tell AI assistants things they wouldn't tell a search bar. Health worries, money stress, a job hunt their boss doesn't know about. The chat history has quietly become a diary.

OpenAI's new locked chats feature is an admission of that. It lets users tuck chosen conversations behind a 4-digit PIN or Face ID, out of the main history and search. On the surface it's a small settings toggle. Underneath, it's a useful case study for anyone building apps that hold personal data.

A lock that's honest about its limits

The most interesting part of the feature is the list of things it doesn't do. OpenAI spells it out plainly. Locking a chat doesn't add end-to-end encryption or a separate vault. It doesn't change how long the data is kept, or whether it's used to train models. It doesn't wipe anything memory picked up beforehand. And it doesn't override workspace policies or compliance access.

That's rare candour. Most privacy features are marketed on how they feel, not what they do. Here the vendor draws the boundary itself, before users assume more than they're getting.

Two kinds of privacy

Locked chats solve for privacy from the room, not privacy from the platform. The stated use case is screen sharing and shared devices: a colleague glancing over during a demo, a partner borrowing the laptop.

That's a real, everyday problem, and it's one most apps ignore. We tend to treat privacy as a backend question of encryption, retention and consent. But for users, the sharpest moments of exposure are social. A notification on a lock screen. A recent-searches list during a meeting. A shared tablet on the sofa.

The detail shows the thinking. Access ends after a minute in the background on mobile, or when a web tab closes. There's a "Lock now" button. OpenAI even warns that other open devices may still show the chat title until they refresh. This is design for the moment someone hits "share screen".

The forgotten-PIN problem

Here's the hard edge. Forget your PIN and reset it, and every locked chat is permanently deleted. So are files and memories held only behind the old PIN. Support can't recover them, and neither can a workspace admin.

It's a deliberate choice. A lock that support can open on request isn't much of a lock. But it pushes a heavy decision onto a 4-digit code people may set once and rarely use. OpenAI repeats the warning several times in its own guidance, which tells you how seriously it takes the risk.

For product teams, this is the core tension of user-held security. The stronger the protection, the more unforgiving the failure mode. The job isn't to avoid that trade-off. It's to make sure users understand it before they're on the wrong side of it.

Memory gets compartments

The quieter shift is what happens to memory. Locked chats can draw on a user's regular memory. But anything learned inside them stays inside, available to other locked chats and nowhere else. Unlocking a chat later doesn't automatically carry those memories across.

This is the start of something bigger: personalisation with compartments. As apps remember more about us, users will want a say not just in what's stored, but in where it surfaces. The assistant that knows about your medical appointment shouldn't mention it in the chat you're screen-sharing with your manager.

Any app building on personal data, from fitness trackers to finance tools, will face the same question. Not "do we remember this?" but "in which context is it okay to use it?"

What app teams can take from it

• Design for social exposure, not just data breaches. Ask where your app embarrasses users in public: previews, recents, notifications, autoplay.

• Say what a feature doesn't do. Clear limits build more trust than vague reassurance, and they protect you when users test those limits.

• Make irreversible actions feel irreversible. If a reset deletes data for good, say so early and often, not just in the final confirmation.

• Treat context as a privacy setting. Personalisation that ignores context will start to feel like surveillance.

• Keep the feature simple. One PIN, one place, one "Lock now" button. Privacy tools only work if people actually use them.

  
  

The bigger signal

Locked chats won't make headlines. But they show where user expectations are heading. People want apps that know them well and keep quiet about it in the wrong company.

The winners won't be the apps with the most impressive privacy claims. They'll be the ones that understand the moments when users feel exposed, and design for those moments honestly.