Is It Safe to Put Company Data Into ChatGPT?

Personal accounts and business accounts aren't the same

On ChatGPT's personal plans (Free, Plus, and Pro), OpenAI may use your conversations to train its models unless you turn that off. The switch is in Settings, under Data controls, labeled "Improve the model for everyone." Temporary Chat is another option: OpenAI says those chats aren't used for training and aren't kept in your history.

OpenAI's business products, including ChatGPT Business (formerly Team), Enterprise, Edu, and the API, aren't used for training by default, according to OpenAI. They also give an administrator control over who uses the workspace, how long data is kept, and which features are switched on.

Plan names and settings change, so check OpenAI's current policy pages before you rely on any of this, including this article.

Not training on your data isn't the same as not keeping it

Even when your chats aren't used for training, they're still stored. Conversations stay in your history until you delete them, and deleted chats can be kept for a period for safety and legal reasons. On a business plan, workspace administrators can generally view and export members' conversations.

That's useful for oversight, but it means a chat is a record. Write in it as if a colleague, an auditor, or a lawyer might read it one day.

What should never go into a chat

Unless a system has been specifically approved for it, keep these out of any AI chat:

• Passwords, API keys, access tokens, and recovery codes.
• Customers' personal details, like ID numbers, payment cards, or health information.
• Anything a contract or NDA says you can't share with a third party.
• Unreleased financial results, deal terms, or plans for acquisitions and layoffs.
• Employee records, performance reviews, and HR investigations.
• Proprietary source code, unless your company has approved a tool for it.

A simple traffic-light rule for your team

Most people don't need a policy document. They need a rule they can remember:

• Green, use freely: public information, your own drafts, and general questions.
• Yellow, use only on the approved business account: internal documents with no personal or confidential client data.
• Red, never: everything on the list above, unless a specific system has been approved for it.

Connectors and agents raise the stakes

ChatGPT can now connect to your email, files, calendar, and other apps, and agent features can browse websites and take actions for you. OpenAI itself warns that once connected, the agent can reach sensitive data in those sources.

The risk is called prompt injection: hidden instructions inside a web page, email, or document that try to trick the AI into doing something you didn't ask for, like sending data somewhere it shouldn't go. A few habits keep that risk small:

• Only turn on the connectors a task needs, and turn them off afterward.
• Use read-only access wherever you can.
• Have a person approve anything that sends, deletes, buys, or shares.
• Let an administrator decide which connectors are allowed for the whole team.

Watch your shared links

When you share a conversation, anyone with the link can read it. In 2025, some shared conversations that people had made discoverable showed up in search results, and OpenAI removed that option. Review the shared links on your account from time to time and delete the ones you no longer need.

Regulated data needs more than settings

If you handle health records, payment cards, or data covered by privacy laws or client contracts, a settings change isn't enough. You need the right agreement with the vendor, a configuration approved for that kind of data, and a written record of what you decided. Don't assume a tool is approved for regulated data because it's popular.

The real risk is shadow AI

If your company hasn't chosen an AI tool, your team has probably chosen one for you, usually on personal accounts you can't see or control. Banning AI rarely works. It mostly pushes the same use out of sight.

The safer approach is to give people a good, approved option and clear rules, so the easy path is also the safe one.

A one-week plan to make AI use safer

• Ask your team which AI tools they already use, and make it clear nobody's in trouble.
• Choose an approved tool and a business plan, with single sign-on if you have it.
• Have an administrator set data retention and turn connectors off by default.
• Write the traffic-light rule on one page and share it.
• Run a one-hour session on what to put in, what to keep out, and how to check the output.
• Put a date on the calendar to review it again in three months.

Where we help

Andrew Wise, our CISO, helps businesses choose approved AI tools, set them up safely, and write rules their teams will actually follow. If you're not sure what's safe to put into AI at your company, that's a good place to start.