Is it safe to paste sensitive data into ChatGPT?
Short answer: for your own, harmless text, usually yes. For other people's personal data, client secrets or credentials, treat the chat like an email to a company you have no contract with. The text leaves your device, and what happens to it then depends on a provider's settings and terms, not on you.
What happens to a prompt
- It is sent to the provider's servers and processed there, often outside your country.
- It may be stored, in your chat history and in the provider's logs, for a period its terms define.
- It may be reviewed or used for training, depending on the product, your plan and your settings. Business plans and API access usually exclude training by default; free consumer accounts often don't unless you opt out.
- It can reach other places: shared chat links, browser extensions with access to the page, screenshots in a ticket.
Check the current terms of the service you use; they change, and they differ between the consumer app, team plans and the API.
Personal data and the GDPR
If the text contains personal data about other people (customers, colleagues, patients, applicants) and you work for an organization in the EU or UK, pasting it into an AI chat is processing under the GDPR. It needs a legal basis, an agreement with the provider when it acts on your behalf, and care with transfers outside the EU. Health data, union membership and other special categories (Article 9) are stricter still. Many employers settle this with an AI policy that lists approved tools and forbids personal data in all others.
Removing the personal data before it is sent is the simplest way to stay on the safe side: what never reaches the provider needs none of the above. Placeholders are pseudonymization, though, not anonymization; see pseudonymization vs. anonymization.
What should never go into a prompt
- Credentials: passwords, API keys, tokens, private keys. Rotate any that were pasted somewhere.
- Payment and identity numbers: card numbers, IBANs, social security and tax numbers, passport numbers.
- Special-category data about identifiable people: health, religion, sexual orientation, union membership.
- Confidential business information under an NDA: unreleased figures, client names, deal terms.
Keep the question, drop the data
The AI rarely needs the real values. A reminder to "[PERSON_1] about invoice [TERM_1]" gets the same help as one with the real name. The routine: paste the text into a tool that runs in your browser, review what it found, replace it with placeholders, ask your question, and put the originals back into the answer locally. The steps are in how to anonymize data for ChatGPT.
This page explains the risks in general terms; it is not legal advice. Your data protection officer can say what your organization allows.