Question & answer

What should you not share with ChatGPT?

The short answer

Keep out anything that would hurt you or someone else if it leaked or landed in training data: your Social Security number, medical records, client data, passwords and contracts. OpenAI may train on personal-plan chats unless you opt out. Strip names and numbers before you paste, or use a tool built for sensitive work.

In detail

Two facts shape this list. On ChatGPT's personal plans, OpenAI may train its models on your conversations unless you opt out, and it retains certain data from your chats either way. Opting out covers new chats only; older ones stay in your history until you delete them.

Your Social Security number. The FTC lists it among the details identity thieves use to open accounts in your name, steal your tax refund or get medical care. No task in ChatGPT needs the real number, so type a placeholder such as [SSN] instead.

Medical records. Lab results, diagnoses and prescriptions reveal more about you than almost anything else you could paste. Need help reading a lab report? Remove your name, date of birth and patient number first, and take the questions that matter to your doctor.

Client data. A customer list, a client's email thread or their account numbers belong to the people in them. In a personal account they fall under the personal rules, training included. For client work, use a business plan such as ChatGPT Business or Enterprise, which OpenAI does not train on by default.

Passwords, API keys and recovery codes. ChatGPT never needs them to help you, and a chat history is a poor place for secrets: whoever gets into your account can scroll back and read them. Keep them in a password manager and describe the problem without the key itself.

Contracts and other confidential documents. Check the confidentiality clause before you paste anything, because sharing the text with an AI company means sharing it with an outside party. If you need help with one clause, paste only that clause and replace names and amounts with placeholders.

For a one-time question, Temporary Chat helps: it is kept off your chat list and is not used for training, although OpenAI may hold on to it for up to 30 days for safety checks (per its help pages on 22 September 2026).

For sensitive work you do every week, an assistant that never trains on chats is the simpler route. Proton's Lumo is one example: Proton says your chats never feed its models, and stored conversations are locked with encryption that Proton itself cannot open.

Five things to keep out of ChatGPT

WhatWhy it stays outDo this instead
Social Security numberIdentity thieves use it to open accounts and claim tax refunds (FTC)Type a placeholder such as [SSN]
Medical recordsDiagnoses and lab results are among the most personal data you haveRemove name, birth date and patient number first
Client dataPersonal plans train on chats unless you opt outUse ChatGPT Business or Enterprise, or anonymize
Passwords and keysAnyone who gets into your account can read your historyKeep them in a password manager
ContractsPasting shares the text with an outside companyPaste one clause, with names and amounts replaced
Head-to-headProton Lumo vs ChatGPT, side by sideLumo and ChatGPT will both draft an email, sum up a document or answer a question, but they handle what you type in very different ways. Here is what each company does with your chats, what each assistant can do, and what the paid plans cost.

Relevant to this question

Lumo logo
No chat training
Lumo

Proton AG

Private AI and encrypted apps4.4
From €9.99/moVisit Lumo