AI and privacy
What ChatGPT stores, who can access it, how to turn off training, and what temporary chats, memory and shared links do. Includes a practical checklist.
For everyday questions without sensitive data, ChatGPT is safe enough if you check its settings. Everything you type or upload goes to OpenAI and stays there until you delete it. On a personal account, it can be used for training. Confidential information about you or other people therefore does not belong there.
This article explores one question; the full overview is on AI chatbots and the GDPR.
People searching for “is ChatGPT safe” usually mean three things at once. Can someone else access my conversations? What does the company behind it do with what I type? And can I trust the answers? This article covers the first two: ChatGPT privacy in practice, and the settings you control. The facts come from OpenAI's privacy policy and help centre as they stood on 13 October 2026. OpenAI regularly changes these pages: open the links before relying on them.
This is an explanation, not legal advice. If you use ChatGPT at work, also read the section on personal and work use.
More than just your questions. OpenAI's European privacy policy, updated on 24 August 2026, lists these types of data.
For Free and Go, the same policy says OpenAI may also collect information about the advertisements you see and click. OpenAI asks for consent for personalised advertising.
OpenAI describes what happens on its page about chat and file retention, as at October 2026:
You cannot recover a deleted chat, including through customer support.
The privacy policy names several groups to whom data may be disclosed:
For users in the EEA and Switzerland, OpenAI Ireland Limited is the controller. The Irish supervisory authority is the lead supervisory authority. According to the policy, the data itself is processed and stored on servers outside the EEA, including in the United States. OpenAI says it relies on European Commission adequacy decisions and standard contractual clauses for this.
If you belong to a ChatGPT Business or Enterprise workspace, the privacy policy says its administrators may access your account and content. If you create an account using your work email address, OpenAI may tell your employer that the account exists.
The greatest chance of a stranger reading your conversations lies with your own sign-in, rather than OpenAI. Anyone with your password sees your entire history. OpenAI offers multi-factor authentication: an authenticator app, a notification on your device, an SMS code or a passkey. Enable it under Settings, Security and login. “Log out of all devices” is in the same place.
On plans for individuals, OpenAI can use your content to train models. Its explanation of how data is used to improve model performance says so. OpenAI also says it takes steps to reduce the amount of personal information in training data.
The setting is called “Improve the model for everyone”. The page about data controls in ChatGPT gives these instructions:
When you are signed in, the choice applies to your account on all your devices.
A temporary chat does not appear in your history, creates no memories and is not used to improve models. Start one by choosing “Temporary” in a new chat.
There are two things to know. OpenAI may keep a copy for up to thirty days for safety: temporary therefore does not mean nothing is stored. Before your first message, you also choose between “Personalized” and “Unpersonalized”. The first option lets the chat use your existing memories and custom instructions. If you do not want that, choose “Unpersonalized”. You cannot change that choice afterwards.
Saving a temporary chat turns it into an ordinary chat, with the settings that apply to your account.
With memory, ChatGPT remembers preferences and details from conversations so you do not have to repeat them. This is convenient, and it means something you mention once may return later.
Manage it under Settings, Personalization, Memory. OpenAI's explanation contains three things you might not expect:
A shared link lets someone else read a conversation of yours. From a personal account, anyone with the link can see it. You cannot restrict a link to particular people or set an expiry date. OpenAI says the pages are not intended to be indexed by search engines, but this does not make a link private: anyone who has it can forward it.
Find the list of your links under Settings, Data controls, Shared links, Manage. Delete them individually or all at once there.
As at October 2026, for a personal account. The names are in English, as OpenAI uses them in its help centre; a Dutch-language app may translate them.
| Setting | What it does | Where to find it |
|---|---|---|
| Improve the model for everyone | Off: new conversations are no longer used to train models | Settings, Data controls |
| Temporary | A chat outside your history, without new memories or training; a copy may remain for up to 30 days | In a new chat, before your first message |
| Unpersonalized | The temporary chat uses no memories or custom instructions | When starting a temporary chat |
| Memory | Determines whether ChatGPT remembers and uses details from conversations | Settings, Personalization, Memory |
| Shared links | Overview of shared conversations, with deletion per link or all at once | Settings, Data controls, Shared links, Manage |
| Delete account | Permanently deletes your account and its associated data | Settings |
| Multi-factor authentication | An extra step when signing in | Settings, Security and login |
Allow ten minutes.
None of these steps changes the starting point: what you type is held by OpenAI.
OpenAI writes in its privacy policy that no transmission over the internet is fully secure, and that you should therefore take particular care over the information you give its services. Its explanation of shared links is more specific: no health information, financial data, passwords or account numbers in content you share. That is also a sound rule for what you type.
OpenAI itself says ChatGPT produces answers by predicting which words are likely to follow, and that they are not always factually correct. Do not blindly trust an answer about your health, money or rights. If an answer contains incorrect information about you, you can ask OpenAI to correct or delete it. The GDPR also gives you rights of access and erasure of your data (Articles 15 and 17 of the GDPR).
As soon as you enter data about clients, patients or colleagues, your organisation processes personal data with a supplier. The GDPR requires an agreement with that supplier for this (Article 28). A personal account, including a paid one, has no such agreement. Your employer also cannot see how your settings are configured.
The Autoriteit Persoonsgegevens (AP), the Dutch data protection authority, received several reports of data breaches caused by staff sharing personal data with an AI chatbot. When this happens on an employee's own initiative and against the agreements with their employer, the AP says it constitutes a data breach. What to do then is explained in Data breach through an AI chatbot: when must you notify the AP?.
For ChatGPT Business and Enterprise, OpenAI's Enterprise privacy page says it does not train models on data by default, and the organisation can enter into a data processing agreement. In return, administrators of a Business workspace can view and export users' conversations. It is therefore not a private space.
Which plan arranges what is covered in ChatGPT Free, Plus, Business or Enterprise: which plan may you use for business?. Find the same comparison for Claude in Claude AI and the GDPR: is Claude safe for business use?. What data may and may not be entered, by type, is covered in May you put client data in ChatGPT, Claude or Mistral Vibe?.
Even with a business plan, one thing stays the same: uploading a case file sends the entire file to the provider. The explanation is on ChatGPT, Claude, Mistral Vibe and the GDPR.
PrivaSearch is not intended for personal use and changes none of ChatGPT's settings. It is made for organisations that want an AI assistant to search client files without the personal data going along.
Documents are analysed inside PrivaSearch. Detected data such as names, BSNs (Dutch citizen service numbers) and addresses are replaced with pseudonyms, and a colleague checks the result before a document becomes searchable. A connected assistant then receives only released passages with a short source reference, and no files. See How PrivaSearch works.
You can connect Claude, ChatGPT Business and Mistral Vibe (formerly Le Chat), through the open MCP standard. A personal ChatGPT account is not included. What the assistant receives from PrivaSearch becomes part of the conversation and is subject to the provider's retention policy. What you type yourself goes straight to the provider. Do not type names in your question even then.
Pseudonymising is not anonymising: for your own organisation, it remains personal data. Agreements with staff, the record of processing activities and the assessment of your own situation therefore remain your responsibility. More about what can go wrong is on The risks of AI for your client files.
To see how this works in practice, request a demo.
For everyday questions without sensitive data, yes, provided you use a strong password with an extra sign-in step and check the settings. Everything you type or upload goes to OpenAI and stays there until you delete it. On a personal account, it can be used for training until you switch that off. Do not enter passwords, medical information or other people's data.
Open the account menu, choose Settings and then Data controls, and switch off “Improve the model for everyone”. When signed in, the choice applies on all your devices. New conversations are then no longer used to train models. Your existing chats remain. If you give a thumbs up or down, that individual conversation may still be used.
OpenAI's privacy policy says data may be disclosed to vendors working for OpenAI, affiliates and, where the law requires it, government authorities. OpenAI says it also monitors content to prevent fraud and misuse. It does not detail precisely who can inspect a personal account's conversation and when. Therefore assume a conversation is not confidential.
The chat disappears from your account immediately, and OpenAI schedules permanent deletion from its systems within thirty days. This does not apply if the chat has already been disassociated from your account, or if OpenAI must keep it longer for security or a legal obligation. You separately delete any memory saved from the chat and any file in the library.
Not entirely. A temporary chat does not appear in your history, creates no memories and is not used for training. OpenAI may still keep a copy for up to thirty days for safety. Choose “Unpersonalized” when starting it if you do not want the chat to use your existing memories. The text goes to OpenAI just as in an ordinary chat.
That depends on the account. On a personal account, ChatGPT arranges nothing for your employer: there is no agreement and your employer cannot see your settings. On ChatGPT Business and Enterprise, your input is not used for training by default and the organisation can enter into a data processing agreement. Your employer decides what you may enter. Ask before sharing client data.
Where an uploaded file ends up, how long it stays, who can access it and what deletion means. With four approaches for files containing personal data.
How schools use AI, why pupil data receives extra protection, which applications are prohibited or high risk, and what a school needs to arrange.
Claude, Microsoft Copilot, Gemini and Mistral Vibe alongside ChatGPT Business: training, agreements, data location and administration. October 2026.
We show it with fictitious client files from your sector: upload, detection, review, search and the connection with your AI assistant. Afterwards you'll know exactly what your team can use safely.
Book a demo →