OpenAI has started rolling out Locked Chats in ChatGPT: pick a conversation, lock it with a four-digit PIN or Face ID or Touch ID, and it moves into its own section, out of your history, pinned chats and search. The feature was reported by TechCrunch on October 20, 2026. It is a useful privacy control for shared screens and shared devices, and it is easy to over-read, because it is not encryption and does not change how OpenAI keeps your data.
This post covers how it works, which chats it excludes, how the separate memory behaves, and the six limits worth knowing before you put anything sensitive behind a lock. We could not find an OpenAI help-center article for the feature at the time of writing, so the details below are TechCrunch's account; check the in-app setting for the final word.
TL;DR: what Locked Chats does
| Question | Answer |
|---|---|
| How do I lock a chat? | Lock it with a four-digit PIN, or Face ID / Touch ID on supported devices |
| Where do locked chats go? | A dedicated Locked Chats section, separate from history, pins and search |
| Is it the same as Temporary Chat? | No. Temporary Chats do not appear in history; locked chats stay and are reachable anytime |
| Does memory carry over? | Locked Chats has its own memory; it can be updated by locked chats but not your regular Memory |
| Is it end-to-end encrypted? | No |
| Does it change retention or training? | No |
| What if I reset the PIN? | All locked chats, files and protected content are permanently deleted |
| Who gets it? | Free, Go, Plus, Pro, Business, Enterprise, Edu; not ChatGPT for Healthcare |
Why OpenAI built it: the shared-screen problem
People use chatbots for the most personal questions they have: health worries, money, relationships, job searches, gift planning. The practical risk is rarely a hacker. It is a partner glancing at a phone, a child handed a tablet, a coworker at your desk, or a screen share where the sidebar shows chat titles.
The feature had been anticipated: a TechRadar headline said hidden code suggested ChatGPT might soon let you lock sensitive chats behind a PIN or fingerprint scan. The release follows a pattern we have seen with other consumer privacy controls: first a temporary mode, then a way to keep selected history private on the device.
TechCrunch describes the intended use plainly: keeping personal conversations or surprise planning private when you hand your device to friends or family.
How Locked Chats works
Based on TechCrunch's account:
- Lock a conversation and set a four-digit PIN, or use Face ID or Touch ID where your device supports it.
- The chat leaves your normal sidebar and search results and appears under a separate Locked Chats area.
- You open it with the PIN or biometric check, and it stays available anytime, unlike a Temporary Chat which is not saved to history.
- If Memory is enabled, the Locked Chats area keeps its own memory, discussed below.
What cannot be locked
TechCrunch lists chat types that do not support locking: Temporary Chats, archived chats, group chats, chats created from a shared link, chats with scheduled tasks, and chats with meeting recordings. If you rely on scheduled tasks or recordings with sensitive content, locking is not an option for those conversations, so they stay in normal history.
The separate memory: the clever part
The detail with the most practical effect is memory. Normally, ChatGPT can carry facts from one chat into later ones. With Locked Chats, new messages in locked chats can update the Locked Chats memory but not your regular Memory, so a sensitive detail does not later surface in an unrelated conversation.
That solves a real leak: you ask about a medical issue in a locked chat, and weeks later a regular chat does not casually reference it while someone is looking over your shoulder. The reverse direction matters too. TechCrunch notes the lock does not remove anything ChatGPT remembered before you locked the chat, so if you moved a conversation into a lock after sharing details, the earlier memory entries are still in regular Memory. Review them in settings and delete what you do not want kept.
Six things Locked Chats does not do
- No end-to-end encryption. There is no separate encrypted vault. The lock controls access in the app, not what happens on OpenAI's servers.
- No change to retention. How long chats are kept is unchanged.
- No change to training settings. The "Improve the model for everyone" setting still applies the same way. If you do not want a chat used for improvement, switch that setting off; do not assume locking does it.
- No removal of earlier memory. Anything remembered before locking stays.
- Not a defense against account compromise. If someone controls your logged-in session or can pass the device's biometric prompt, the lock is only as strong as that gate. A four-digit PIN has only 10,000 combinations.
- Deleting by resetting the PIN. The reset path permanently deletes locked content, which protects privacy but means a forgotten PIN equals lost chats.
If you need a conversation not to exist at all afterwards, a Temporary Chat or deleting the conversation is a different tool with different tradeoffs. For the broader trust question, see our reporting on how OpenAI's human review of chats works under Project Lily.
Locked Chats vs Temporary Chat vs deleting
| Tool | Appears in history? | Keeps memory? | Good for |
|---|---|---|---|
| Normal chat | Yes | Updates regular Memory | Everyday work |
| Locked chat | In the Locked Chats section only | Own separate memory | Private topics on a shared screen or device |
| Temporary Chat | No | Not saved to history | One-off questions you do not want kept in history |
| Delete chat | Removed | Memory entries may remain until you delete them | Getting rid of a conversation |
The right mental model: Locked Chats is a door lock inside your house, not a safe. It keeps casual onlookers out. It does not stop the landlord, who is the provider, from having a key under its own policy.
What this means for teams and businesses
The feature is listed for Business, Enterprise and Edu plans, but admins should not assume it changes compliance posture. Retention, export and eligibility for review are governed by workspace policy rather than a client-side lock, and a lock on an employee's device does not give employers or auditors anything new. If your organization restricts what employees may paste into AI tools, a lock is not a substitute for that policy. Our guides on data breaches involving AI agents and API keys and ChatGPT Dots sending an unapproved email show the kinds of failures where a screen lock does not help: the risk is in what an agent or service does with data, not who sees your screen.
A note on ads and personalization
OpenAI has been expanding monetization in ChatGPT, including visual ads tests and reports of custom audiences targeted by email or phone. Those stories are why many users now ask what a chat feature does and does not change about how their conversations are used. For Locked Chats, nothing in the reporting says locked content is treated differently for ads or training; the training toggle is the control to check. We have not seen an OpenAI statement on ad personalization for locked content, so do not assume either way.
How to use it well
- Lock first, then type. Start a sensitive conversation inside the lock instead of moving it later, so earlier memory is not created.
- Pick a biometric where you can. Face ID or Touch ID is harder to shoulder-surf than a four-digit PIN.
- Review regular Memory and remove anything sensitive that was saved before you used the lock.
- Check the training toggle if you also want the conversation excluded from model improvement.
- Write the PIN down somewhere safe because a reset deletes everything.
- Do not store secrets you cannot lose. Passwords, API keys and legal documents do not belong in a chatbot regardless of the lock.
Who should use it, and who should not
Locked Chats fits three everyday situations well. The first is a shared device: a family tablet or a laptop passed around at home, where you do not want a health, money or job-search conversation visible in the sidebar. The second is screen sharing, where chat titles in history can show up in a recording or a video call. The third is surprise planning, the use case TechCrunch itself highlights, where a gift or event chat should stay hidden from whoever picks up your phone.
It fits poorly when the real concern is the provider, an employer, or a legal process. In those cases the lock changes nothing, because retention, training settings and workspace policy sit outside it. It also cannot cover chat types that do not support locking, such as group chats, shared-link chats, and chats with scheduled tasks or meeting recordings. For those, the options are the normal history controls, a Temporary Chat, or deleting the conversation.
What is still unknown
- The final help-center wording, which may differ from the press account.
- Whether administrators on Enterprise and Edu plans can disable the feature.
- Exactly how locked content is treated in data exports and legal holds.
- Whether the web, desktop and mobile apps all got the feature at the same time. The rollout began Tuesday and may not be visible to every user yet.
Related reading
- OpenAI Project Lily: human review and ChatGPT privacy
- ChatGPT Dots and the unapproved email
- ChatGPT visual ads test
- ChatGPT ads custom audiences: claimed vs verified
- What is a data breach: types, AI agents and API keys
- WhatsApp help pages with hidden LLM instructions
Details are as reported on October 20, 2026 and may change as OpenAI's rollout completes.