AI and privacy: what never to share with a chatbot
AI and privacy: what happens to your conversations, the 7 things never to share with a chatbot and the privacy settings worth checking today.
We talk to an AI as we would to a confidant. It’s available, patient and never judges, which makes it handy and sometimes leads us to say a little too much. The data you give an AI tool deserves your attention. A few habits reduce the risks; they go hand in hand with choosing a suitable tool and settings. They let you enjoy AI without letting slip anything that should stay yours.
What happens to your conversations with an AI?
When you write to an AI assistant, your message goes to the servers of the company that provides it. What happens next depends on the service and your settings.
- Your conversations may be saved in a history, so you can find them again.
- They may be kept for a certain time, even after you delete them on your side, depending on the service’s terms.
- In some cases, people may review them to improve quality or safety.
- Depending on the service, the plan and your settings, your conversations may be used to improve or train models. Check the terms that apply to your account.
The rules vary a lot from one service to another, and between free and business versions. The simplest approach is to assume that what you write could one day be read by someone else.
7 things never to share with a chatbot
Some information has no place in a conversation with an AI, however much you trust the service.
- Your passwords and codes, including those for your bank or your email.
- Your card numbers and account details.
- Your identity numbers, such as your National Insurance number, passport, driving licence or tax reference.
- Detailed health information that identifies you, especially alongside your name.
- The private information of people close to you, such as their health or contact details, without their agreement.
- Confidential work documents, such as contracts, customer files or internal figures. Only share them in an approved tool, for an approved use and according to your organisation’s rules.
- Anything you wouldn’t tell a stranger, for example your full address alongside the times you’re away, or details about your children.
Settings worth checking
Take five minutes to open your assistant’s settings. Most major services offer privacy options, and four of them deserve a look.
- The history setting tells you whether your conversations are kept and whether you can delete them.
- The training setting tells you whether your conversations are used to improve the models and whether you can opt out.
- Depending on the tool and the options chosen, a share link can make a conversation visible to anyone who receives it. Check the permissions before passing it on.
- Connections to your apps give the AI access to your email, calendar or files. Keep only the ones you actually use.
Settings and their names vary from one tool and one plan to another. The official pages of the main assistants explain them.
- For ChatGPT, OpenAI’s Data Controls FAQ.
- For Gemini, the Gemini Apps Privacy Hub.
- For Claude, “Is my data used for model training?” in Anthropic’s Privacy Center.
- For Mistral’s Vibe, Mistral AI’s privacy policy.
How to enjoy AI without oversharing
You can almost always get help without handing over sensitive information. Here are a few tricks.
- Replace names with roles and remove details that could identify people. To practise, prefer a made-up example.
- Remove numbers, addresses and references from a document before pasting it.
- Describe the situation without the details. “A letter from my insurer refusing a claim” is usually enough.
- Reread before sending, as you would an important email.
These habits take a few seconds and save plenty of regrets. To go further, find out how to check AI answers before relying on them.
Cautious or trusting: each profile has its own habits
On these questions, we don’t all start from the same place. The Guardian always asks themselves what they can and can’t share, and their caution helps a lot here. The Architect can rely on explicit rules and review them as tools and uses change. More trusting profiles gain from giving themselves two or three simple rules, so they no longer have to think about it.
Our test asks for no account and no name, and your answers stay in your browser, as explained in our privacy promise. In about 3 minutes, it offers a reading of your relationship with AI and with information. Take the free test to discover your profile and a first tip suited to you.