Safe AI use: 7 rules for your privacy
What you type into a chatbot goes to a company’s servers. These seven rules help you use AI safely, without needless privacy risks.
An AI chatbot can feel like a private conversation, but everything you type goes to a company’s servers. If you want to use AI safely, it helps to know what you should and should not share. These seven rules help you get the best from AI, without needless risks for yourself, your clients or your employer.
Why does privacy matter when you use AI?
What you type or upload in a chatbot leaves your computer and ends up with the company behind the tool. It may be stored there, reviewed for quality control or used to improve the AI. That depends on your settings.
This is not just theory. The Dutch Data Protection Authority received several reports of data breaches in which staff had put customer or patient data into an AI chatbot, reports Security.NL. The regulator warns that in effect you are sharing that data with a tech company, without the people involved knowing.
New to this kind of tool? Then first read what AI is and how it works. That will also help you understand why you need to be careful about what you share.
Safe AI use: what are the 7 rules?
The basic idea is simple: only share what you would also feel comfortable putting in a public place, and check your settings. These seven rules will help.
- No passwords or codes. Never type passwords, PINs or login codes into a chatbot.
- No identity details. Your national insurance number, passport number or bank account number do not belong there.
- No medical details about others. If you ask about health, keep it general and do not name people.
- No confidential work information. Customer data, contracts and internal figures should only be shared in a tool approved by your employer.
- Make data anonymous. Replace names with customer A or colleague B before you paste in text.
- Check your settings. Many tools have a setting that lets you choose whether your conversations may be used to train the AI, and whether the tool remembers things about you.
- Watch out for fakes. Only download AI apps from the official maker. Fake apps and fake websites try to steal data or money.
What can you and can’t you share with a chatbot?
General questions and anonymised text are usually fine. Personal and confidential data are not. Use this table to choose quickly.
| Fine to share | Better not, or only anonymously | Never share |
|---|---|---|
| General questions, such as an explanation of a concept | An email to a customer, with names replaced | Passwords, PINs and login codes |
| A public text you want shortened | A meeting summary, without names and details | National insurance number, passport number or account number |
| Ideas for a presentation or social post | A question about health, without names | Medical data about others, contracts and customer files |
Want to know how to ask good questions without sensitive details? In our article about what a prompt is you can read how to write a clear instruction.
How do you change your privacy settings?
In most AI tools, you will find an option in the settings to choose whether your conversations may be used to train the AI. Turn that off if you do not want that.
In ChatGPT, this option is called Improve the model for everyone. You will find it in the settings under Data Controls. According to OpenAI’s explanation, your new conversations are no longer used to train models if you switch this option off. Also check whether the tool has a memory feature and what it stores about you.
Using ChatGPT for the first time? Our guide for beginners will help you step by step.
How do you spot deepfakes and AI scams?
If you get an urgent request for money, even from someone you know, always check it through another channel. For example, ring that person yourself on a number you already had.
With AI, scammers can make messages, voices and videos that look increasingly real. A familiar voice on the phone is no longer proof. If in doubt, ask a question that only the two of you would know. Have you still become a victim? Then report it to the Fraud Help Desk and file a police report, advises Radar by AVROTROS.
How do you anonymise a text with help from AI?
Remove the names yourself first, then let AI check whether any recognisable details are still in the text. That way, you never share the real data.
Frequently asked questions
Does anyone read what I type into a chatbot?
They might. Depending on the tool and your settings, conversations may be stored and sometimes reviewed for quality control. So assume that what you type is not private.
Is a business version of an AI tool safer?
Often, yes. With business versions, you can usually make agreements about how data is processed. Still, always read the terms and discuss it with your employer or privacy lead.
What should I do if I have accidentally shared sensitive data?
Delete the conversation. Have you shared a password? Change it straight away. Was it customer or colleague data from work? Report it to your employer immediately, because it may be a data breach.
May I use AI for my work?
That depends on the rules in your organisation. Ask which tools are approved and what data you are allowed to put into them.
Your next step
Open the settings of the AI tool you use most today and check whether your conversations are used for training. Want to go further after that? Read how to start using AI at work safely and smartly.
Was this article helpful?
