Is It Safe to Put Company Data Into ChatGPT at Work?

Only in a tool and account your company has approved for that kind of data. On personal plans (ChatGPT Free, Plus or Pro, for example), what you type is stored by the vendor outside your company's control, and it may be used to train future models unless you turn that off. Business plans such as ChatGPT Business or Enterprise are not used for training by default, and your company controls who can see the chats and how long they're kept. Your company's policy and IT team decide what is allowed, not the vendor's settings page.

• The account matters more than the brand. "Copilot" or "Gemini" with a personal login and with a work login come with different data terms.

• "Not used for training" doesn't mean "not stored". Vendors keep chats for history and safety checks, and company admins may be able to read them.

• Some things never go into any chat: passwords, API keys and access codes.

• Placeholders let you get help without the details. Write [CUSTOMER NAME] and fill it in yourself.

• If you pasted something by mistake, tell IT quickly. Deleting the chat alone may not be enough.

When you send a prompt, it travels to the vendor's servers, where the model writes the answer. What happens next depends on the product and your settings:

You send a prompt (plus any files you attach)
        │
        ▼
Vendor's servers: the model writes the answer
        │
        ├── Chat history: kept so you can reopen it, until you delete it
        │   or your company's retention setting removes it
        ├── Safety checks: copies kept for a limited time to detect abuse
        ├── Human review: some vendors let trained reviewers read a sample
        └── Model training: on personal plans often allowed unless you
            turn it off; on business plans off by default

Two terms come up on every vendor page:

• Training means using conversations to improve future versions of the model. Your content becomes part of the material the vendor learns from.

• Retention means how long the vendor keeps your data. Deleting a chat usually removes it from your view at once and from the vendor's systems after a set number of days.

Feedback buttons are a detail people miss. On several personal plans, clicking thumbs up or thumbs down sends the whole conversation to the vendor, even if you turned training off.

The table below summarizes each vendor's own help and privacy pages, checked on October 20, 2026. These terms change often, so open the linked page before you rely on a detail. The update date is the one each page showed.

Three things in that table matter most for everyday work:

• Business plans are designed for company data; personal plans are not. The vendors themselves make this split. Google's personal Gemini page asks you not to enter confidential information that you wouldn't want a reviewer to see.

• Your company can see business chats. OpenAI says ChatGPT Business admins can view, export and delete members' conversations, and Microsoft says IT can view logged Copilot prompts. That is a feature for the company, not a privacy gap, but it means a work chat is not a private diary.

• Names changed in 2026. Microsoft renamed Microsoft 365 Copilot to Microsoft Copilot, so the same name now covers personal and work use. Check which account you're signed in with, not which logo you see.

Most companies sort data into levels, such as public, internal, confidential and restricted. Use your company's levels if it has them. If it doesn't, this rough guide is a reasonable starting point while you ask IT:

Some uses of AI are regulated in some places. Ask your legal and compliance team what applies to you before you roll this out to a team. And whatever the tool, AI should not make decisions about people on its own. Hiring, performance, pay, credit and health decisions need HR or legal involved and a person responsible for the outcome.

Unless IT has approved a tool for this kind of data, keep these out of AI chats:

• Credentials: passwords, API keys, access codes, recovery codes. Never, in any tool.

• Personal details about customers or colleagues: names together with contact details, account or ID numbers, health information, salaries, performance reviews.

• Unpublished numbers: financial results before they're public, prices under negotiation, deal terms.

• Contracts and legal documents, including anything connected to a dispute.

• Product secrets: unreleased plans, designs and source code.

• Anything a client gave you in confidence, or that your company labels confidential or restricted.

Search your intranet for "AI policy" or "generative AI", or ask your manager. Look for two lists: which tools are approved, and which data you may use in each. If your company has no policy yet, ask IT before you use AI with any work data.

The same app can run under personal or business terms depending on your login. Sign in with your work account, and look for your company's name in the account menu or workspace switcher. If you have both accounts on one device, keep work chats in the work account.

Where your company allows personal accounts for non-confidential tasks, check the data settings first:

• ChatGPT: Settings > Data controls > Improve the model for everyone. Temporary chats aren't used for training.

• Claude: Settings > Privacy > Help improve our AI models. Incognito chats aren't used for training.

• Gemini: turn off Keep Activity in Gemini Apps Activity, or use a temporary chat.

• Microsoft Copilot: in Settings > Personalization, decide on memory and Allow ads personalization.

Changing a setting doesn't turn a personal account into an approved work tool. It only reduces what the vendor does with your chats.

You can get most of the help without the sensitive part. Instead of pasting a customer's name, email, card number and invoice, describe the situation and use placeholders:

Task: Draft a reply to a customer who says we charged her twice for the same invoice.

Context: We checked and she is right. The duplicate payment of [AMOUNT] will be refunded to her card within 5 business days. She will get a confirmation email when the refund is sent.

Audience and tone: She is upset. Apologize once, be plain and specific.

Format: Under 100 words.

Rules: Use the placeholders [CUSTOMER NAME], [INVOICE NUMBER] and [AMOUNT] exactly as written. I will fill them in myself.

The real answer from OpenAI's gpt-5-mini:

Hi [CUSTOMER NAME], we checked and you’re right — invoice [INVOICE NUMBER] was charged twice. I’m sorry for the error. The duplicate payment of [AMOUNT] will be refunded to your card within 5 business days, and you’ll receive a confirmation email when the refund is sent.

You paste the real name, invoice and amount into your email program, not into the chat. ChatGPT, Claude, Gemini and Copilot will word the reply differently, so check that every placeholder survived before you fill them in. For more on writing prompts like this, see AI prompts for work.

Removing names isn't always enough. A rare job title, an exact date and a small office can identify a person together. If the situation itself is sensitive, ask before you use AI at all.

• Uploaded files follow the same rules as pasted text. A spreadsheet export often holds more than you meant to share.

• Connectors that link the assistant to your email, calendar or shared drives can let it read anything you can open in those services. Use only the ones IT has turned on.

• Browser extensions and AI meeting note-takers can read pages or record whole meetings. Many companies have separate rules for these. Ask before you install one.

• Documents and web pages can carry hidden instructions that try to steer the AI. This is called prompt injection; prompt injection explained shows how it works.

Safe input doesn't guarantee a correct output. AI tools still get names, numbers and dates wrong. Check what matters before you send or decide anything; how to check AI answers gives a five-step check.

Act quickly and don't hide it:

1. Stop and don't add more to that chat.

2. Delete the chat. The vendors above remove deleted chats from their systems after a set time, often 30 days, so deleting limits the exposure but doesn't undo it.

3. Tell your manager or IT security, following your company's incident process.

4. If you shared a password, key or access code, change it at once.

Copy this list into an email or a chat with your IT or security team.

1. Which AI tools are approved for work, and which account should I sign in with?
2. Which kinds of data may I use in each tool, by our data classification?
3. Is our workspace set so that our chats aren't used to train models? Do I need to change any setting myself?
4. How long are chats kept, and who in the company can read them?
5. May I upload files, or connect email, calendar or shared drives?
6. Are browser extensions, plugins and AI meeting note-takers allowed?
7. Who do I tell if I paste something I shouldn't have?
8. Who approves new uses, such as AI with customer data or anything that affects decisions about people?

Does ChatGPT keep what I type?

Yes. Regular chats stay in your history until you delete them, and OpenAI schedules deleted chats for permanent deletion within 30 days. Temporary chats don't appear in your history, but OpenAI may keep a copy for up to 30 days for safety. In ChatGPT Business and Enterprise, your company's admins set how long chats are kept.

If I turn off training, is my data private?

It's more private, not private. Turning off "Improve the model for everyone" stops OpenAI from using new chats to train models. The chats are still stored, can be checked for abuse, and follow the retention rules above. In a company workspace, your admins may also be able to view them. Paste only what your company allows.

Is ChatGPT Plus safer for work than the free version?

Not for data. Free, Plus and Pro are all plans for individuals and share the same data terms: OpenAI may use your content for training unless you opt out. ChatGPT Business, Enterprise and Edu are the plans with business data terms. The same split exists for Claude (Free, Pro, Max vs Team and Enterprise).

Can my employer see my ChatGPT chats?

In a company workspace, often yes. OpenAI says ChatGPT Business admins can view, export and delete members' conversations, and Enterprise admins can access an audit log. Microsoft says prompts in Copilot with a work account are logged and IT can view them. Your employer can't see a personal account, which is one reason personal accounts aren't the place for work data.

Is Microsoft Copilot safer than ChatGPT for company data?

It depends on the account, not the brand. Copilot signed in with a work account comes with Microsoft's enterprise data protection. ChatGPT Enterprise comes with OpenAI's business terms. Either can be the right tool if your company approved it. Copilot with a personal Microsoft account is a consumer product, like ChatGPT Free.

Who decides what I'm allowed to put into AI tools?

Your company does, through its policy, its IT and security team and, for some uses, its legal and compliance team. Vendor settings only control what the vendor does with your data. They don't make a use allowed at your company. When in doubt, ask before you paste.

• Write prompts that get usable drafts without sensitive details: AI prompts for work

• Check facts, numbers and dates before you send: how to check AI answers

• Understand hidden instructions in documents and web pages: prompt injection explained

• Learn why AI tools get facts wrong: LLM hallucinations and quick checks

• For leaders planning company-wide use: how to roll out AI to your team

• When AI starts taking actions, not just writing: what is human in the loop AI?

• For IT teams that build AI tools: PII detection and redaction