Are AI chatbots safe? Privacy and what not to share

ChatGPT, Claude, and Gemini are not going to give you a virus or empty your bank account. The safety question is really a privacy question: what do you share with them, where does it get stored, and who might see it later. Once you know the rules, you can use AI without giving up things you do not need to give up.

What happens to your conversations

• Stored on company servers. Every conversation is saved to your account, by default. You can usually see and delete past chats in settings.
• Automated scanning. Companies scan chats for abuse (illegal content, threats, attempts to extract sensitive data) and safety issues. This happens automatically.
• Used for training (sometimes). ChatGPT free uses your chats to improve future models, unless you turn this off. Gemini does too. Claude does not, by default.
• Human review. Companies say employees only look at conversations in narrow cases: when flagged, when investigating abuse, or when you submit feedback. Either way, treat chats as semi-private, not strictly private.

Who actually sees your chats?

In practice:

1. You. Anyone with access to your account can see your chat history. Use a strong password and turn on two-factor authentication.
2. Automated systems. Every message passes through abuse-detection systems. These are computers, not people.
3. The company, in rare cases. Engineers debugging an issue, safety teams investigating abuse, or in response to a subpoena.
4. Other users? No. Your chats are not shared with other accounts. The "ChatGPT shared a stranger's chat" stories from 2023 were bugs that got fixed.

What NOT to paste into a chatbot

These are the lines to never cross:

• Social Security numbers. Full or last four digits.
• Credit card numbers. Including expiration date or CVV.
• Bank account or routing numbers.
• Passwords. Yours or anyone else's.
• Photos of IDs. Driver's license, passport, Social Security card.
• Login credentials for any service.
• Two-factor authentication codes.
• Medical records with your full name attached.
• Children's full names and birthdays (especially if asking how to set up accounts for them).
• Sensitive work documents that your employer would care about.

What is OK to share

• General questions about how things work
• Drafts of emails (with names changed if needed)
• Documents with sensitive parts removed or replaced with placeholders
• Photos of public things (your dog, plants, products)
• Your address only if you have a reason (e.g., "suggest restaurants near 95076"). A ZIP code is usually enough.
• Your medical situation in general terms (not your full chart)

How to remove personal info before pasting

Easy method: replace identifying details with placeholders before sending.

• "My SSN is 123-45-6789" becomes "my SSN is [XXX]"
• "Pay $1,250 to account 987654321" becomes "pay [amount] to [account]"
• Names: "John Smith" becomes "[client name]" or just "my client"
• Addresses: "456 Oak Street, Watsonville, CA" becomes "my address"

The AI does not need real numbers to help you draft a letter or explain a contract. Placeholders work fine.

Privacy controls for each chatbot

ChatGPT

1. Click your name (bottom left) > Settings.
2. Click Data Controls.
3. Turn off Improve the model for everyone. This stops OpenAI from using your future chats to train models.
4. You can also turn off Chat history & training entirely. New chats will not be saved to your sidebar.
5. Click Export data if you want a copy of all your past chats.
6. Click Delete account at the bottom to wipe everything.

Claude

1. Click your name (bottom left) > Settings.
2. Click Privacy.
3. Anthropic does not train on your conversations by default, so there is less to turn off.
4. You can delete individual chats by clicking the three dots next to them in the sidebar.
5. You can delete your whole account from this menu too.

Gemini

1. Go to myactivity.google.com.
2. Find Gemini Apps Activity.
3. You can pause Gemini activity, set auto-delete (every 3 / 18 / 36 months), or delete past activity.
4. In the Gemini app, click the gear > Activity to access the same controls.

Common privacy worries (and the real answers)

"Will AI use my information to scam me?"

The companies (OpenAI, Anthropic, Google) are big regulated companies. They are not selling your data to scammers. The risk is breaches and subpoenas, which is true of any cloud service. Behave the same way you would with Gmail or Dropbox.

"Can my employer see what I typed into ChatGPT?"

If you use ChatGPT on a work computer or a work account, possibly yes. Companies can install monitoring software. Companies with paid ChatGPT Team or Enterprise plans can see chats from employees on those accounts. If you do not want work to see, use a personal account and a personal device, and do not paste anything your employer would care about.

"Is voice mode listening when I am not talking?"

Voice mode listens while the conversation is active, then stops when you close the app or end the call. Unlike Alexa or Google Home (which listen for a wake word continuously), ChatGPT and Claude voice modes require you to start the conversation. They are not always-on microphones.

"Will AI remember things I said before?"

ChatGPT Plus has a "memory" feature that lets it remember details across chats (your job, your kid's name, your preferences). You can turn this off in Settings > Personalization > Memory. The other chatbots do not have this on by default. If memory is on, treat each chat like the previous one is still loaded.

"Are screen-share videos sent to ChatGPT private?"

ChatGPT's "Vision" feature (uploading images, screen shares) sends those images to OpenAI's servers, same as your text chats. Cover or blur sensitive info before sharing screen with AI.

Watch out for fake AI apps

This is the single most common AI safety risk in practice. There are dozens of fake "ChatGPT" and "AI assistant" apps in the App Store and Play Store. They harvest data, charge subscriptions you cannot cancel, or are scams. Always:

• Only download the official apps. ChatGPT is by OpenAI. Claude is by Anthropic. Gemini is by Google LLC.
• Check the developer name on the store listing.
• If in doubt, install via the link on the company's real website (chatgpt.com, claude.ai, gemini.google.com).
• Read our spot fake apps guide for the full checklist.

Watch out for AI scams

AI is now being used by scammers in new ways:

• Voice cloning scams: 10 seconds of your voice from social media is enough to clone it. Then scammers call your relatives asking for money. Set a family safe word with everyone you care about; if "you" call asking for money, the safe word is required.
• Deepfake video calls: not common yet but possible. If a relative's video call has a weird delay or odd lighting, hang up and call them back directly.
• Phishing emails written by AI: better grammar, more personalized. The old "spot the typo" rule no longer works. See our phishing guide.

The takeaway

Use AI chatbots like you would any cloud service: convenient but not strictly private. Do not paste in identifiers a thief could use. Turn off training where you can. Watch for fake apps and AI-powered scams. Past that, ChatGPT, Claude, and Gemini are safe to use. The benefits genuinely outweigh the risks for most everyday tasks.