Is ChatGPT safe for client files?
For client files, the answer depends on your ChatGPT plan and your rules. As of November 2026, OpenAI says it does not train its models on ChatGPT Business or Enterprise data by default. On personal plans, it may, unless each user turns it off. Anthropic and Microsoft say the same for their business plans. Your firm still needs plain rules and a person who checks every draft.
01 / Plan by plan
Which AI plans train on your firm’s chats?
On personal plans, a vendor may train on your chats. On business plans, the vendors say they do not by default. Training means using your chats to improve the AI.
| Plan | Used for training by default? (November 2026) | Who controls the data settings | Where to check |
|---|---|---|---|
| ChatGPT personal plans, such as Free, Plus and Pro | It may be, says OpenAI, until the user turns it off. Rated chats may still be used. | Each user | OpenAI Help Center |
| ChatGPT Business and Enterprise | No, says OpenAI, unless your firm agrees to share data. | Your workspace admins | OpenAI enterprise privacy |
| Claude personal plans, such as Free, Pro, Max and Ultra | If the user allows it, says Anthropic. Flagged or rated chats may also be used. | Each user | Anthropic Privacy Center |
| Claude Team and Enterprise | No, says Anthropic. Chats sent as feedback may be used. | Your organization’s owners | Anthropic Privacy Center |
| Microsoft Copilot and Copilot Chat, with a work account | No, says Microsoft, for prompts, answers and the files Copilot reads. | Your Microsoft 365 admins | Microsoft Learn |
A workspace is your firm’s shared account. An admin runs it. To compare plans, see which AI plan fits a firm.
02 / ChatGPT
Is ChatGPT confidential?
ChatGPT treats chats in different ways on personal and business plans. As of November 2026, OpenAI’s help page says it may train on chats on personal plans, such as Free, Plus and Pro. Each user can turn this off in Settings, under Data controls, with “Improve the model for everyone.” A thumbs up or down can let OpenAI train on that whole chat, even after the user opts out.
ChatGPT Business and Enterprise
Here is what OpenAI’s enterprise privacy page says about these plans.
- OpenAI does not train on business data by default.
- Each person sees their own chats.
- On Business, your admins can view, export and delete members’ chats. OpenAI staff may look for support, abuse checks or legal reasons, and some contractors for abuse checks.
- On Enterprise, your admins decide how long chats are kept. They can also pull a log of chats with an admin tool. OpenAI staff look only to fix incidents, to recover chats with your permission, or when the law requires it.
- Deleted chats leave OpenAI’s systems within 30 days, with some exceptions.
03 / Claude
Is Claude safe, and does it train on your data?
As of November 2026, Anthropic’s page for business plans says it does not train on Claude Team or Enterprise chats by default. On personal plans, such as Free, Pro, Max and Ultra, Anthropic’s page for personal plans says it trains on chats if the user allows it in Settings, under Privacy. Chats flagged for a safety review may also be used.
Ratings are an exception on every Claude plan. A thumbs up or down saves the whole chat as feedback, which Anthropic keeps up to 5 years and may train on. Team and Enterprise owners can turn ratings off.
Who controls retention
Retention means how long a vendor keeps your data. On Team and Enterprise, Anthropic’s retention page says deleted chats leave its systems within 30 days, with some exceptions. Enterprise owners can also set how long chats are kept.
04 / Microsoft Copilot
Is Microsoft 365 Copilot confidential?
As of November 2026, Microsoft’s Copilot privacy page says Copilot does not use your prompts, its answers or the emails and files it reads to train the large AI models behind Copilot. These terms are for Copilot and Copilot Chat as used by organizations, says Microsoft’s data protection page. Staff should sign in with a work account.
Microsoft says Copilot shows each person only the data they can already open. Prompts and answers are stored under the same contract terms as your firm’s other Microsoft 365 content. Your admins can set how long they are kept. The controls vary by plan.
When Copilot searches the web, it sends a short query to Bing, under separate terms.
05 / What stays out
What should stay out of every AI tool?
Passwords, ID and account numbers, and client work in personal accounts should stay out of every AI tool. A firm can adopt these rules.
- Put client work only into your firm’s business plan, signed in with a work login.
- Never paste passwords, login codes, or full ID, bank, card or tax numbers.
- Take out names and private details when the task does not need them.
- Keep out anything a client asked you to keep out of AI.
- Do not give client chats a thumbs up or down. On some plans, that lets the vendor keep or train on the chat.
Privilege and professional duties
Your firm decides how this fits your own professional duties. Ask your own counsel.
06 / Your AI use rules
What goes on one page of AI use rules?
One page of AI use rules says which tools staff may use, what stays out, who reviews each draft, what to do when the AI is wrong, and who to ask.
| The page says | A made-up example line |
|---|---|
| Which AI tools staff may use | Use only our firm’s AI workspace, with your work login. |
| What never goes into AI | Never paste passwords, or ID, bank, card or tax numbers. |
| Who reviews each draft | A partner reviews every client letter before it goes out. |
| What to do when the AI is wrong | Fix the draft and tell the workflow owner. |
| Who to ask | Ask the office manager when you are not sure. |
07 / How we handle data
How does Ego Eimi handle client data?
We build on test data. Your client data stays with your firm. Your firm owns what we build for it: the workflow, its template and the AI use rules.
We start with one workflow in Claude, ChatGPT or Copilot. A workflow is one task your team does again and again, done the same way with AI. Your staff use the AI under your firm’s own agreement with your AI provider.
See how the 30-day build works and how we handle data and ownership. For your field, read AI for law firms, AI for accounting firms or AI for engineering firms.
08 / Common questions
Is ChatGPT safe for confidential information?
It depends on the plan and your rules. As of November 2026, OpenAI says it does not train on ChatGPT Business or Enterprise data by default.
Does ChatGPT, Claude or Copilot train on our data?
Not by default on business plans, say OpenAI, Anthropic and Microsoft, as of November 2026. On personal plans, OpenAI may train on chats unless the user opts out, and Anthropic trains on them if the user allows it.
Who can see our chats?
Each person sees their own chats, and your admins run the workspace. OpenAI says ChatGPT Business admins can view, export and delete members’ chats. Enterprise admins can pull a log of chats with an admin tool.
What should never go into an AI tool?
Passwords and login codes. Full ID, bank, card or tax numbers. Client work in a personal account. Anything a client asked you to keep out of AI.
Does using AI affect privilege?
Your firm decides how this fits your own professional duties. Ask your own counsel.
What should our AI use rules say?
Which tools staff may use, what never goes in, who reviews each draft, what to do when the AI is wrong, and who to ask.
09 / Sources
- OpenAI Help Center: How your data is used to improve model performance · checked November 20, 2026
- OpenAI: Enterprise privacy at OpenAI · checked November 20, 2026
- Anthropic Privacy Center: Is my data used for model training? (Claude Free, Pro and Max) · checked November 20, 2026
- Anthropic Privacy Center: How do I change my model improvement privacy settings? · checked November 20, 2026
- Anthropic Privacy Center: Is my data used for model training? (Claude for Work) · checked November 20, 2026
- Anthropic Privacy Center: How long do you store my organization’s data? · checked November 20, 2026
- Anthropic Privacy Center: Configure custom data retention controls for Enterprise plans · checked November 20, 2026
- Microsoft Learn: Data, Privacy, and Security for Microsoft Copilot · checked November 20, 2026
- Microsoft Learn: Enterprise data protection in Microsoft Copilot and Microsoft Copilot Chat · checked November 20, 2026
Last updated November 20, 2026.
Next / Your first call
Free · 30 minutes
Start with one task. See the result on day 7.
Before the call, we send you a one-page estimate about your firm. On the call, we update it with your real numbers.