The problem
ChatGPT is increasingly becoming a place where people have conversations that are more personal and sensitive than the conversations they would normally have with other people.
Projects make this even more valuable because they allow long-running conversations to retain context over time.
However, there is currently a privacy gap between account security and conversation privacy.
For example:
I may have ChatGPT open and authenticated on my phone. A family member, friend, or colleague may ask to use my phone to search something. I am comfortable handing them my unlocked phone and letting them use ChatGPT.
But I do not want that person to be able to open a personal Project such as “Mental Health” or read conversations about my private life.
Today, protecting the account does not solve this situation because the account is already unlocked.
Proposed feature: Private Projects / Private Conversations
Allow users to designate a Project or individual conversation as Private.
A Private Project/conversation would have an additional privacy boundary independent of normal ChatGPT account authentication.
For example:
Normal
-
Visible in the normal sidebar
-
Searchable normally
-
Accessible whenever the ChatGPT account is unlocked
Private
-
Hidden from the normal sidebar when locked
-
Excluded from normal chat search
-
Not exposed through conversation suggestions/history
-
Requires an additional authentication step (device biometrics, PIN, or account re-authentication) to access
A particularly useful mode: Guest Mode
Add an optional Guest/Borrow Mode for situations where someone temporarily uses the user’s device.
When enabled:
-
Private Projects disappear from the interface
-
Private conversations are excluded from search
-
Private memories/context are unavailable
-
Private files are hidden
-
The guest can use ChatGPT normally for their own questions
-
Exiting Guest Mode requires biometric/PIN authentication
This solves a very common real-world situation:
“I am willing to let you use my unlocked ChatGPT account, but I am not giving you permission to browse my private conversations.”
Important: privacy should not mean Temporary Chat
The goal is not to force users to use Temporary Chat.
Many users deliberately want persistent context for deeply personal conversations.
A Private Project should therefore remain persistent and retain its context for the owner, while simply being inaccessible to other people using the already-authenticated session.
For example:
A user could have:
-
Work → Normal
-
Travel → Normal
-
Coding → Normal
-
Personal Journal → Private
-
Mental Health → Private
The private conversations would continue to retain their normal project context for the owner.
Security model
The key distinction is:
Account authentication ≠ authorization to view every conversation.
A user can be authenticated to ChatGPT while a particular conversation remains separately protected.
This is similar to how other applications can have sensitive areas that require an additional authentication step even after the device/account has already been unlocked.
Why this matters
The threat model here is not a hacker breaking into an account.
It is a much more ordinary situation:
Someone is physically using a device that the account owner has intentionally unlocked.
The owner may have no problem letting that person use ChatGPT, but they should not have to choose between:
-
refusing to let anyone use their phone, or
-
exposing deeply personal conversations.
A project-level privacy boundary would solve this without sacrificing ChatGPT’s long-term memory and context.
Suggested MVP
A simple first version could be:
-
Mark a Project as Private.
-
Private Projects are hidden from the normal sidebar.
-
Add “Unlock Private Projects.”
-
Require device biometric/PIN or ChatGPT re-authentication.
-
Exclude locked private projects from chat search.
-
Add an optional “Guest Mode” that automatically hides all private projects.
This would already solve the core problem without requiring major changes to the conversational experience.
The broader principle
ChatGPT is becoming a place where users store increasingly intimate and important parts of their lives.
As that happens, users need more granular control than simply:
“Is my account logged in?”
They need:
“Who is authorized to see this particular part of my AI history?”
I believe Projects are the natural place to introduce that privacy boundary.