Skip to content

Knowledge that lights the world

What Should You Never Tell a Chatbot? Privacy and Data Security When Using AI

Where does what you type into AI chatbots go? Eight kinds of information you should never share, chat history and model-training settings, safe use at work and practical ways to anonymise sensitive information.

What Should You Never Tell a Chatbot? Privacy and Data Security When Using AI

AI chatbots have become part of everyday life, from writing emails to understanding health questions, from summarising contracts to fixing code. Because the chat interface feels personal, people sometimes type details into these tools that they would not even tell a friend. Yet every message you write goes to the servers of the company providing the tool and, depending on the service’s settings, may be stored for some time.

In this article we explain what happens to what you type into AI chatbots, which information you should never share and which settings to check to protect your privacy. The aim is not to fear these tools but to use them knowingly.

Where does what you type into a chatbot go?

When you send a message to a cloud-based chatbot, the text is sent to the company’s servers so that a response can be generated. Although it varies by service, three issues usually matter:

  • Retention: Your chat history may be kept in your account until you delete it or for a set period. Deleted chats may also be kept for a while for security and abuse monitoring.
  • Model training: Some services may use user chats to improve their models. This is usually an option you can switch on or off in the settings; the default differs by service and account type.
  • Human review: Privacy policies may state that some chats can be reviewed by a limited number of authorised staff for safety or quality purposes.

Reading the privacy policy and data settings of the tool you use once is therefore the most basic security step.

Eight kinds of information you should never type into a chatbot

  • Passwords and verification codes: No legitimate AI service asks for your password or a one-time code sent by text message.
  • ID, passport and tax numbers: This information can be used for identity theft.
  • Card and bank account details: Even when getting help with a budget, you do not need to type a real card number, IBAN or account details.
  • Detailed health records: You can get help understanding a test result, but remove your name, ID number and hospital details.
  • Other people’s personal data: Do not share the name, phone number or private circumstances of a friend, customer or employee without their permission.
  • Company secrets: Unpublished financial results, customer lists, source code or draft contracts should not be uploaded to general-purpose tools without your employer’s permission.
  • Full address and daily routine: Information such as your home address, your children’s school and when you are away from home creates a security risk.
  • Legally sensitive documents: Documents relating to an ongoing case or investigation should not be shared without your lawyer’s advice.

Settings to check to protect your privacy

Most AI services have similar privacy options in their settings:

  • Chat history: You can turn history off or delete it regularly. Some services offer a “temporary chat” mode that is not saved.
  • Use for model training: If you do not want your chats used to improve the model, turn this option off.
  • Connected apps: If you have given the assistant access to your email, calendar or files, regularly review which permissions are switched on.
  • Account security: Because your chat history sits in your account, use a strong password and two-step verification.

Using AI at work

Many companies introduced rules after incidents in which employees uploaded confidential information to general-purpose AI tools. For safe use at work, first find out your company’s AI policy. Enterprise versions usually guarantee by contract that data is not used for model training and apply stricter access controls. When in doubt, consult your information security team.

A practical way to anonymise sensitive information

To get help, you often do not need real information. When polishing an email, replace names with labels such as “Customer A”; when having a table analysed, delete real names and numbers or replace them with dummy values. If you want to understand a contract clause, share only the relevant clause with the parties’ names removed. This small habit greatly reduces risk without giving up the tool’s benefits.

Also remember that chatbot answers are not always correct. In areas such as health, law and finance, AI answers are not a substitute for professional advice.

Ask yourself before sharing

Pausing for a few seconds to ask these questions before typing information into a chatbot prevents most risks. This checklist is especially useful when you are working in a hurry or copying and pasting long documents.

  • Would it be a problem if this information were public? If the answer is yes, remove or generalise it.
  • Does this information belong to someone else? To share another person’s data you need their permission and a legitimate reason.
  • Do I really need real data? Often you can get the same help with sample or dummy data.

If children or teenagers in your family use AI, talk to them about these rules too. Young people may not always realise the risks of sharing information such as their school name, address and photos; reviewing account settings together is a good start.

Frequently asked questions

Do deleted chats disappear completely?

Deleted chats are removed from your account, but many services may keep them for a while for security and legal obligations. You can find the period in the service’s privacy policy.

Can I stop my chats being used for model training?

Many services offer this option in their settings. In enterprise versions, data is usually not used for training by default.

Is it safe to ask an AI a health question?

It can be used for general information, but do not share details that identify you, and always see a doctor for diagnosis and treatment.

Are models running locally on a computer more private?

With models running on your own device, data usually does not leave the device, so they can be a more private option for sensitive tasks. However, the security of your device still matters.

Sources and further reading

  1. European Data Protection Board (EDPB), opinion on AI models and personal data, December 2024 (edpb.europa.eu)
  2. US Federal Trade Commission (FTC), warnings on AI and consumer privacy (ftc.gov)
  3. Turkish Personal Data Protection Authority (KVKK), “Guide on Generative AI and the Protection of Personal Data” guide (kvkk.gov.tr)
  4. OWASP, “Top 10 for Large Language Model Applications” (owasp.org)
Share
Follow YulaHubGet the day's most important stories first.
RSS

Leave a comment

Your email address will not be published. Required fields are marked *