6 min read
If you paste a client’s contract into ChatGPT and then lock that chat behind a PIN, is the contract now private? Most people would say yes. The honest answer is: private from whom? That question is the whole point of this comparison.
On October 20, 2026, OpenAI started rolling out Locked Chats in ChatGPT, which hides selected conversations behind a four digit PIN or Face ID. It is a genuinely useful feature. It is also easy to mistake for something it is not. So let us put it side by side with the other “private” modes solo owners already have, and test each one against the risks that actually matter for client work.
The three risks you are really trying to manage
Before comparing buttons, name the threats. For a freelancer or small business owner handling client information, there are three:
- The shoulder risk: someone with access to your laptop or phone (a family member, a coworker at the shared desk, whoever borrows your device) scrolls through your chat history.
- The memory risk: the assistant remembers details from a sensitive chat and brings them up later, possibly in a conversation you screen share with a different client.
- The provider risk: the text is stored on the AI company’s servers, may be reviewed for safety, and may be used to improve future models depending on your settings.
Every privacy feature covers some of these. None covers all three. Keep that in mind as we go.
ChatGPT Locked Chats: strong against the shoulder risk
What it does, according to coverage of the launch from TechCrunch and BetaNews:
- Locked conversations disappear from your regular history, pinned chats, and search, and live in a separate “Locked Chats” section.
- You open them with a four digit PIN, or with Face ID or Touch ID on supported devices.
- If Memory is on, the locked section gets its own separate memory, so details from locked chats do not leak into your everyday conversations.
- It is rolling out across Free, Go, Plus, Pro, Business, Enterprise, and Edu plans, gradually, so you may not see it yet.
What it does not do: OpenAI says plainly that locking a chat does not add end-to-end encryption (a setup where only you hold the key) and does not change how long data is retained. The conversation still sits on OpenAI’s servers under the same rules as before.
One trap worth underlining: if you reset your PIN, the locked chats are permanently deleted, and support cannot recover them. Do not store the only copy of anything important in there.
Verdict: excellent against the shoulder risk, good against the memory risk, no change to the provider risk.
ChatGPT Temporary Chat: strong against memory, weak for anything you need later
Temporary Chat has been around longer. A temporary conversation does not appear in your history, does not read from or write to Memory, and OpenAI says it is not used to train its models. OpenAI may still keep a copy for up to 30 days for safety review before deleting it.
The catch is practical, not technical: once you close it, it is gone from your side. That is perfect for “rewrite this sensitive email once” and useless for an ongoing client project.
Verdict: strong against the shoulder and memory risks for one-off tasks, modest improvement on the provider risk.
Claude Incognito: the same idea, on Anthropic’s side
Claude’s incognito mode, marked by a small ghost icon when you start a new chat, works much like ChatGPT’s Temporary Chat. Incognito conversations are not saved to your chat history and are not used for Claude’s memory across chats. Anthropic still retains them for a limited period under its safety policy, so the provider risk is reduced rather than removed.
Verdict: nearly identical to Temporary Chat. Choose based on which assistant you already use, not on this feature.
Gemini Temporary Chats: the shortest storage window
Google’s Gemini app offers temporary chats too. They are not shown in your history, not used to personalize future answers, and not used to train models, and Google says they are kept for up to 72 hours. If a short storage window is your priority among the big three consumer apps, this is the shortest one. As our free Gemini vs ChatGPT vs Claude comparison notes, though, the free tier now runs a lighter model, so the quality trade-off is real.
The side-by-side table
- ChatGPT Locked Chats: hidden from device snoopers (yes), separate memory (yes), kept for later (yes), changes provider storage (no).
- ChatGPT Temporary Chat: hidden (yes), no memory (yes), kept for later (no), provider storage (reduced, up to 30 days).
- Claude Incognito: hidden (yes), no memory (yes), kept for later (no), provider storage (reduced, limited safety window).
- Gemini Temporary Chat: hidden (yes), no personalization (yes), kept for later (no), provider storage (reduced, up to 72 hours).
Notice the pattern. Locked Chats is the only option that keeps the conversation for later. The other three are about forgetting. That makes them complements, not competitors.
What actually reduces the provider risk
If the real worry is “my client’s data is on someone else’s server,” no hide button solves it. These do:
- Turn off model training in settings. In ChatGPT it is the “Improve the model for everyone” toggle under Data Controls. Claude and Gemini have equivalent switches. Check them today; defaults change.
- Use a business plan when clients require it. Business and team tiers typically exclude your data from training by default and come with clearer contracts. If a client’s agreement mentions confidentiality, this is the tier that matches it.
- Remove identifying details before pasting. Replace names, addresses, and account numbers with placeholders like [CLIENT] or [ADDRESS]. The AI rarely needs them to help.
- Keep the most sensitive work off the cloud entirely. For documents you truly should not upload, running a small model on your own computer is now realistic, as our guide to running a private AI model on your laptop walks through.
A simple rule set for client work
Here is how I would combine these features in a one-person business:
- Ongoing client projects with mild sensitivity (draft proposals, meeting notes): regular chats or Projects, training off, client names replaced. Lock the chat if your device is shared.
- One-off sensitive tasks (rewriting a tense email about a dispute): Temporary Chat, Incognito, or Gemini’s temporary chat.
- Personal or business matters you want hidden but kept (your own finances, a health question, a hiring decision): Locked Chats, with the PIN stored in your password manager.
- Anything covered by an NDA, medical, legal, or financial account details: a business plan with a clear data agreement, or a local model, or no AI at all.
Write that list into your AI policy so you are not deciding from scratch every time. If you need a template, start with our one-page AI policy quick wins.
The skeptic’s bottom line
Locked Chats is a good feature that solves a real, narrow problem: people with access to your device. It does not make ChatGPT a vault, and OpenAI does not claim it does. Hiding a conversation and protecting a conversation are different jobs. Use the lock for the first, and your settings, your plan, and your redaction habits for the second.
Which of these modes do you actually use day to day, and has a client ever asked you how you handle their data in AI tools? Share your answer in the comments.
