AI tools can be useful, but use them carefully: share less sensitive information, limit access, and verify important results.
1. Don’t share secrets with AI chatbots
Avoid entering passwords, recovery codes, credit card numbers, ID documents, confidential work documents, or other sensitive information into AI services.
If you wouldn’t share something with a regular online service, don’t share it with an AI service either.
2. Don’t upload files without thinking
Photos, PDFs, screenshots, and spreadsheets can contain personal information, metadata, or other confidential data.
Before uploading a file, ask yourself:
Does this contain anything I wouldn’t want an AI service to have?
3. Check your privacy and data settings
Before using an AI service for sensitive matters, check its privacy settings.
Pay attention to:
- Whether your conversations can be used to improve the service.
- What conversation history and memory settings do.
- How long your data is stored.
- Whether your data can be shared with connected applications.
If you don’t need a setting that affects your privacy, consider turning it off.
4. Don’t give browser extensions unnecessary permissions
AI-powered browser extensions may be able to access the websites you visit or information you enter.
Only install extensions you need, and check the permissions they request.
5. Don’t automatically trust AI-generated links, files, or instructions
AI can generate misleading links, recommend fake software, or create phishing messages.
Check important instructions, links, and files against a trusted source before using them.
6. Don’t rely on AI alone for important decisions
AI can make mistakes or provide outdated information.
For important matters such as banking, legal issues, health, or security settings, verify key information using an official source or consult a qualified professional.
7. Don’t trust AI-generated voices, videos, or messages
AI can make it easier to convincingly impersonate someone else.
If someone urgently asks for money, passwords, verification codes, or sensitive information, verify the request through a separate, trusted channel.
8. Don’t give AI unnecessary access
Only give AI the permissions it actually needs.
For example:
- An AI tool that summarizes emails may not need permission to send or delete messages.
- An AI tool working with one file may not need access to all your files.
- An AI tool that checks your calendar for available times may not need permission to change calendar entries.
- If AI doesn’t need your camera, microphone, or contacts, don’t give it those permissions.
The more an AI tool can access or control, the greater the potential impact if something goes wrong.
9. Don’t trust instructions found in external content
AI can receive information from websites, emails, files, or connected applications. That content may contain instructions designed to influence what the AI does. This is called prompt injection.
For example, a webpage might contain text telling the AI to do something different from the summary the user requested.
Check that a link is trustworthy before asking AI to open it or summarize it.
The risk is greater when AI can take actions on your behalf, such as sending messages or editing files.