Is my AI chat private? The short answer
Your AI chat is private from people who pick up your phone only if you lock or hide it. It is not private from the company that runs the tool, and it becomes visible to anyone who gets a share link you create. “Private” is really five separate questions.
A lock hides a chat from the person holding your phone. That's all it does. The company, the training setting and the deleted-chat clock are separate.
This is for anyone who has typed something into ChatGPT, Gemini, Claude or Meta AI that they would never text to a group chat. It covers personal accounts only. A work account is different, because your employer's admin may see more.
The grid below shows all five, and a 5-minute checklist closes the guide. For the bigger picture, see the best AI assistant for your personal life. Updated October 22, 2026.
Who can actually see your AI chats?
Five audiences matter, and none gets ranked by odds here. What matters is that every fix has a blind spot.
| Who | Can they see it? | What protects you | What does NOT protect you |
|---|---|---|---|
| Someone holding your phone | Yes, if the chat is in your history | A phone passcode, an app lock such as ChatGPT's Locked Chats, or a temporary or incognito chat | A lock does not change what the company stores |
| Anyone with a share link | Yes, no invite needed | Not creating the link, or deleting it | Deleting a shared chat can leave the public link alive (Gemini) |
| The company and its reviewers | In principle, under its policy | Reading the vendor's privacy page | Locks, incognito and deleting do not remove the company's own access |
| Model training | Possibly, depending on settings | The training opt-out in the app | Turning it off does not pull back data already used |
| Deleted-chat retention | Kept for a while after you delete | Nothing you can toggle | The delete button hides a chat before it erases it |
Sources: TechCrunch (Oct 20, 2026), WIRED (Oct 21, 2026), CNET (updated Oct 1, 2026), and OpenAI, Google and Anthropic help pages.
What did ChatGPT's new Locked Chats actually change?
On October 20, 2026, TechCrunch reported that ChatGPT users “can now lock ChatGPT conversations to keep sensitive chats private when others may have access to their screen or device.” You secure them with a four-digit PIN, or Face ID or Touch ID on supported devices.
A screen lock, and a handy one. If a partner or kid borrows your phone, or the family tablet gets passed around, a locked chat stays shut.
The rollout began October 20 and, per TechCrunch, “may not be live yet for all users.” If you don't see a lock option in the app, it probably just hasn't reached you yet.
What Locked Chats does NOT protect
Per TechCrunch's report of OpenAI's own materials:
- It is not end-to-end encryption. There is no separate encrypted vault either.
- It does not change how long OpenAI retains the chat.
- It does not remove what ChatGPT remembered before you locked the chat.
- It does not change your model-training setting.
- It does not work for every chat type. Temporary, archived, group and shared-link chats are left out.
- Resetting the PIN is destructive.It “permanently deletes all locked chats, files, and other content protected by the previous PIN.”
Who can see a chat when you share a link?
A share link is a web address that opens your chat for whoever holds it. No account, no invite. Anyone who finds or forwards it can read it too.
The freshest example is Meta's Muse app. Per WIRED (Oct 21, 2026), researchers at the security firm Claroc “were able to enumerate 6,800 links Muse users created for sharing conversations with others.” Among the few hundred they opened were a small-business owner's loan application and a to-do list with a bank fraud alert and a partial account number.
Share links to Muse conversations that researchers were able to enumerate, no invite needed.
The cause: the conversation title sat in the web address's subdomain. As of WIRED's Oct 21 report, sharing was paused: Meta told The Verge it had “paused this functionality while we review this report and consider changes to how these links can be discovered.” Paused is not removed, so check the app before you assume it is off. More on using Meta AI as a personal assistant.
To be fair, users chose to create those links. And this isn't just one app. The Verge notes that ChatGPT, Claude and Grok conversations have wound up on Google Search before.
Google's own help page adds a catch for Gemini: deleting a shared chat does not delete its public link. Look for “Your Public Links” in Gemini's settings to manage them (Google, Oct 13, 2026).
Free habits that help: don't share chats with money, health or family details, remove the link when you're done, or send a cropped screenshot instead.
Can the company read your chats, and is training on by default?
In principle, yes. The company stores your chat, and under its policy staff and reviewers can see content.
Google is blunt about it. Its Gemini help pages say “a subset of chats are reviewed by human reviewers (including Google's trained service providers),” and warn: “Please don't enter confidential information that you wouldn't want a reviewer to see or Google to use to improve our services.” For ChatGPT and Claude, human review isn't covered in the sources reviewed here, so check each privacy page.
Training
CNET (updated Oct 1, 2026) reports that “by default, most chatbots will use your conversation data for training purposes, but if you know where to look, you can usually turn this option off.”
Where to find the switch, from vendor pages:
- ChatGPT:look for Settings > Data controls > “Improve the model for everyone.” OpenAI says turning it off “does not delete or hide saved chats,” and thumbs up or down feedback can let a whole conversation be used.
- Gemini:look for Settings & help > Activity, “Keep Activity” (Google help, Oct 2026).
- Claude:Anthropic's page says that after you switch training off, data already in training runs or in models already trained is not removed.
- Meta AI:CNET wrote it “was unable to find the specific setting” to stop Meta using your data, and Meta did not respond. Meta cites controls like wiping memories or disconnecting services, per WIRED (Oct 3, 2026).
Training is its own setting. Locking a chat does not turn it off.
Memory
ChatGPT's “Dreaming” memory, per Tom's Guide (June 18, 2026), builds a profile from your past conversations automatically. OpenAI acknowledged the memory summary page “may not capture everything ChatGPT remembers about you.” It was a staged rollout, so it may not be on your account yet. Tom's Guide placed the review screen under Settings > Personalization > Memory, but menus move around.
What happens when you delete a chat?
Deleted doesn't mean gone right away. Here's what is documented.
- ChatGPT (deleted chats):CNET reports OpenAI “also stores deleted chats for up to 30 days.”
- ChatGPT (Temporary Chat): OpenAI's help page says it “may keep a copy of a temporary chat for up to 30 days for safety purposes.” Temporary chats stay out of history.
- Claude:deleted conversations leave your history immediately and are deleted from Anthropic's back-end storage within 30 days (Anthropic, page dated July 1, 2026). Incognito chats (ghost icon, top right) are not saved to history or memory.
- Gemini: activity auto-deletes after 18 months by default. Temporary chats and chats with Keep Activity off are kept for 72 hours. Chats that human reviewers have looked at can be kept up to three years and are not deleted when you delete your activity (Google, Oct 2026).
- Meta AI: deletion and retention details are not documented in the sources reviewed here.
Delete can't undo two things: a memory made from a chat can outlive it, and training that already happened stays happened (CNET, Anthropic).
How to delete properly
- Delete the chat.
- Check memory for anything the chat left behind.
- Check for shared links and remove them.
- Check the training setting.
A 5-minute privacy checklist for your phone
In order of payoff. Open the app and do these now.
- Turn on a lock.Use Locked Chats in ChatGPT if you see it, or your phone's own app lock. It only hides chats from the person holding the phone.
- Switch off training where the app lets you (see the paths above).
- Review memory.
- Find and remove old share links.Gemini's “Your Public Links” is the one first-party path with a documented home.
- Use a temporary or incognito chat for the awkward questions. It stays out of history, but the company can still see it.
- Keep account numbers, health details and family details out of chats you might share. Treat that as a habit, not professional advice.
If you already shared something you regret
Delete the link. Delete the chat. Then contact the company that runs the tool and ask what it can do.
Related: signs you are relying on AI too much. You may meet Muse first: Sensor Tower estimates about 10.9 million downloads since its September debut (third-party estimate via TechCrunch, Oct 20, 2026; downloads are not users).
Frequently asked questions
Can someone see my ChatGPT chats if they use my phone?
Yes, if your chat history is visible when they open the app. A phone passcode and ChatGPT's Locked Chats, where available, can hide chats from whoever holds your device. Temporary Chats do not appear in history. None of these stop the company from storing the chat.
Does locking a chat make it private from OpenAI?
No. Locked Chats is a screen lock, not encryption. TechCrunch reports it does not change how long OpenAI retains the chat, remove earlier memories, or change the training setting. It only keeps people with your screen or device out of that chat.
Are AI chat share links public?
Effectively, yes. Anyone who has the link can open the chat, no invite needed. Researchers found about 6,800 Muse links through public DNS records, per WIRED (Oct 21, 2026). The Verge notes ChatGPT, Claude and Grok shares have also appeared on Google Search. Only share what you would be fine seeing public.
Does deleting an AI chat actually delete it?
Not immediately. CNET reports OpenAI stores deleted chats for up to 30 days, and Anthropic says Claude deletes from back-end storage within 30 days. Google says reviewed Gemini chats can be kept up to three years. Deleting also does not undo training that already happened.
Can I stop an AI chatbot from training on my chats?
Often, yes, going forward. CNET says most major chatbots let you turn training off, but it will not remove data already used in training. For Meta AI, CNET could not find a specific setting and Meta did not respond. Check each app's settings and help page.
Is a work AI account different from a personal one?
Yes. A work account can be visible to your employer's admin, so it follows different rules. This guide covers personal accounts only. For a work login, ask your employer what is visible before you type anything personal into it.
Want more plain-language guides? Join the newsletter, or head back to what a personal AI assistant can do.