What you must not tell or ask ChatGPT: 10 forbidden topics for your own safety

1. Passwords and account login details

This happens more often than you might think: someone asks for help setting up a service and, in passing, includes their password directly in the message. ChatGPT does not encrypt the information entered and cannot guarantee that it will not fall into the wrong hands — which is why passwords and usernames should only be stored in dedicated password managers (Bitwarden, 1Password, KeePass), and certainly not in a chatbot.

2. Personal documents: passport, tax identification number (TIN)

It sometimes happens that someone wants to check whether their documents have been filled in correctly and uploads a scan or text containing real details to the chat. The TIN, passport series and number, and date of birth — all of these are personal details which, in the hands of fraudsters, can be used to take out loans or forge documents.

ChatGPT doesn’t need your real details to explain how to fill in a form correctly. Simply type «replace your real tax identification number with XXX» — and the explanation will be just as clear, without any risk to you.

3. Financial details: card numbers, account numbers, CVV

Your bank card number, account details and CVV code are all the information someone needs to make a payment in your name. Entering this information in any chat, including ChatGPT, is like leaving your wallet open on a park bench.

Even if you think you’re just checking that the details are correct — don’t do it. There are secure channels for such tasks: online banking and the bank’s official support service.

4. Investment advice and financial decisions

«Where should I invest my money?», «Is it worth buying shares in X?», «Which cryptocurrency will rise in value?» — ChatGPT can provide an answer to these questions, but that answer will carry no legal or financial liability.

AI does not know your actual financial situation, your risk appetite, your goals or your time horizon. It can explain how the market works — but it should not give specific investment advice, and it is dangerous to rely on it too heavily. Financial decisions require a licensed financial adviser, not a chatbot.

5. Medical diagnoses and health data

Asking ChatGPT, «What does this symptom mean?» is fine. But if you enter your test results, details of chronic conditions or the medicines you’re taking, that’s a different matter altogether.

Firstly, this is sensitive personal information that may remain in the database. Secondly, medical diagnoses require a doctor — a specialist who sees you as a whole person, knows your medical history and is accountable for their actions. ChatGPT may give an alarming or incorrect result, and this is no joke: there have been cases where people have genuinely suffered as a result of chatbots’ misinterpreted «diagnoses».

6. Trade secrets and employment NDAs

Imagine that your company is developing a new product and you want ChatGPT to help you prepare a presentation. You enter internal figures, plans and client names into the chat. It’s quick and convenient, but beyond your control.

If you’ve signed a non-disclosure agreement (NDA), sharing such data with a chatbot almost certainly breaches it. Several companies (notably Samsung) have already experienced actual leaks of corporate information via ChatGPT. For working with confidential material, there are corporate solutions with their own security measures, rather than public chatbots.

7. Personal addresses, telephone numbers and location

Your home address, telephone number and places you frequently visit are details which, when combined with other information, can be dangerous. Fraud, phishing and social engineering all start with precisely these «little things».

ChatGPT does not need your address to answer a question about a neighbourhood or town. For geolocation purposes, the name of the town or street is sufficient, without a specific house number.

8. Illegal and unethical requests

What can ChatGPT do — and what should it not do? The chatbot has built-in filters, but some users try to circumvent them: they ask it to describe criminal schemes, the synthesis of dangerous substances, ways to hack accounts, or other harmful information under the guise of ‘educational’ requests.

The problem here is not just that the chatbot might (albeit rarely) respond to such requests. The problem is also that the requests themselves are logged. OpenAI (the company that created ChatGPT) can analyse suspicious sessions and pass them on to the relevant authorities in accordance with legal requirements. Illegal requests pose a risk not only to you, but to everyone around you.

9. Advice during difficult personal and psychological crises

There are times when people turn to a chatbot in search of support — following a tragic loss, or whilst experiencing stress or depression. ChatGPT may seem like a sensitive and attentive conversational partner, but it cannot replace a real person.

Firstly, it bears no responsibility for the consequences of its words. Secondly, in difficult situations, it may say something that is technically correct but emotionally damaging. Thirdly, having someone nearby — a psychologist, a helpline, or a loved one — is always a better choice than a text-based algorithm.

10. Requests to write manipulative or fraudulent content

«Write a convincing letter, supposedly from a bank» or «draft a review that sounds natural» — such requests may seem harmless, but in reality they are asking you to create a tool for manipulation or fraud. ChatGPT has filters, but phrasing such as «for training» or «as an example» sometimes gets round them.

The problem here is twofold: firstly, such content can genuinely harm other people — victims of phishing, fake reviews or psychological pressure. Secondly, the request is logged, and if its purpose is obvious, this already poses a legal risk to you personally. Manipulating people using AI is not a grey area, but a clear-cut violation.

How to write safe prompts

If you want to use ChatGPT effectively and safely, here are a few simple rules to follow when creating prompts for it:

• Replace real data with placeholders. Instead of your tax identification number, write «Tax ID: XXXXXX»; instead of your real address, write «Town N, Street Y». The chatbot will understand the task without any personal details.
• Do not paste entire corporate documents into the chat. Describe the task in your own words — this is usually sufficient.
• Check your privacy settings. In ChatGPT’s settings, you can disable the use of your chats for model training. To do this, click on your profile in the bottom-left corner of the screen, open «Settings» → «Data Controls» and disable the «Improve the model for everyone» option. After that, new chats will not be used to train OpenAI models.
  
  
• Think of ChatGPT as a clever stranger It can help you find your way, but you wouldn’t give it your home address or leave your wallet with it.

And on home security

When discussing digital security, it’s worth mentioning where exactly these conversations take place. If your home Internet connection is unreliable or unsecured, the risks increase even without ChatGPT. A vulnerable router, an open network, or an unstable connection provide an easy point of entry for those seeking to intercept your data.

A stable home Internet connection isn’t just a matter of convenience; it’s the foundation of digital security. An unreliable, intermittent connection forces you to use open networks in cafés or shopping centres, where the risk of data interception is significantly higher. Maxnet’s Internet service provides a seamless home connection with technical support available to answer any relevant queries. Fewer disconnections mean fewer temptations to go online via unsecured access points.

Digital literacy isn’t paranoia. It’s the habit of treating your data with the same care as your house keys: don’t leave them lying around and don’t trust just anyone.