Keeping Your Private Data Safe When Using AI

AI chats feel private, like talking to yourself. They are not. What you type goes to a company's servers, may be stored, and under some settings may be reviewed by people or used to improve products. A few habits keep you safe without giving up the benefits.

Know where your text goes

Different products and plan types handle data differently. Things that commonly vary:

• Whether your conversations are used to train future models
• How long chats are kept, and whether you can delete them
• Whether staff may review conversations for safety or quality
• Whether business or enterprise accounts have separate terms

Do not assume. Open the privacy or data controls in your account and read the vendor's current policy. It takes five minutes and the answer can differ by product, region and plan.

Never paste these

• Passwords, API keys, recovery codes and private keys
• Financial account and card numbers
• Government ID numbers
• Medical records tied to a named person
• Other people's personal details without a good reason
• Confidential company material you are not authorized to share

If you need help with a config file or log, replace secrets first. A key pasted into a chat should be treated as exposed, so rotate it.

Redact before you paste

Swap real details for placeholders:

• Names become [CLIENT A], [PERSON 1]
• Companies become [COMPANY]
• Account numbers become [ACCOUNT]
• Addresses and phone numbers become [REDACTED]

You can even ask an assistant to help find sensitive items in a document, but only do that with a tool and plan you trust for that data. For truly sensitive material, do the redaction yourself.

Using AI at work

Follow your employer's rules. If there is no policy, ask before using customer or company data. Practical guidance:

• Use the company-approved tool and account, not a personal one, for work material.
• Keep client data, source code, financials and unpublished plans out of unapproved tools.
• Treat AI output as a draft that you are responsible for.
• Tell your manager how you use it. Transparency avoids trouble later.

Settings worth checking

• Training and improvement opt-outs: many products let you turn off use of your chats for training.
• Chat history and memory: decide what the assistant may remember about you, and clear it when needed.
• Temporary or incognito chats: some assistants offer chats that are not saved to history. Read what that mode actually does.
• Connected apps: when you link email, drive or calendar, the assistant can read those. Grant the minimum access.
• Sharing links: shared chat links can be seen by anyone who has them. Share carefully and delete links you no longer need.

Files, images and voice

Documents, screenshots and recordings may contain more than you notice: names in a margin, location data in a photo, a client list in a hidden sheet. Check before upload.

Browser extensions and third-party apps

Many apps claim to add AI to your tools. Each is another company handling your text. Prefer well-known providers, review permissions, and avoid extensions that ask to read every site you visit.

A quick pre-paste test

Ask yourself:

1. Would I be comfortable if this appeared in a data breach?
2. Am I allowed to share it?
3. Can I remove the sensitive parts and still get the help I need?

If any answer is no, redact, change tools or skip the task.

Local models

Running an open model on your own computer keeps text on your machine. It is a fair option for sensitive work if your hardware can handle it, though local models can be weaker than the largest hosted ones and still need sensible security.

Summary

Check settings, redact by default, use approved tools at work, and never paste secrets. Good habits make AI both safer and easier to defend.