Start · Simple
Is company data safe in AI tools? What to check before staff paste in documents
Staff already paste emails, offers and tables into AI tools, whether the company has a rule or not. Whether that is safe is not a matter of opinion: it is written in the terms of the plan they use.
Photo: Nilo Velez, public domain (CC0)
In this guide
Short answer
It depends on the plan and its terms, and you can check them yourself. Five points decide it: whether your inputs are used for training, where data is stored, who at the vendor can see it, how long it is kept, and whether a data processing agreement exists. Some things should never be pasted in, whatever the plan.
Is it safe to use AI with company data?
It is as safe as the terms of the exact plan your staff use, and those terms differ from plan to plan of the same product. Free and consumer plans often have different rules from business plans, above all on whether what you type may be used to improve the vendor’s models. So the name of the tool tells you little. The plan does.
There are two separate risks. The first is what the vendor does with your text: store it, let employees read it, train on it. The second is what your own staff put in: a text that should never have left the company is not made safe by good terms. The first risk is settled by reading; the second by a rule everyone knows.
What should you check in the terms of your plan?
Check five points, and get each answer from the vendor’s own terms or privacy documents, not from a comparison site.
| Point | Question to answer | A good answer |
|---|---|---|
| Training | Are our inputs and uploads used to train or improve models? | No, or only if we switch it on ourselves |
| Storage location | In which country or region is our data stored and processed? | A named region that fits your obligations |
| Access at the vendor | Who can read our conversations, and when? | Limited, logged access for stated reasons |
| Retention | How long are conversations and files kept, and can we delete them? | A stated period, and deletion on request |
| Data processing agreement | Does the vendor sign one for this plan? | Yes, available for the plan you use |
Write the answers down with the date and the name of the plan. Terms change, and “we checked this in spring” is worth something only if someone knows what was checked.
Two further points matter once more than a few people use the tool. Can the company manage the accounts centrally, so that a leaver loses access on the last day? And are the conversations of one employee invisible to the others unless shared on purpose?
When do you need a data processing agreement?
You need one whenever the tool processes personal data on your behalf. Under the GDPR, a company that lets a service provider process personal data needs a contract with that provider covering the processing. Names, email addresses, customer correspondence and staff records are all personal data.
In practice this means: if the plan your staff use offers no such agreement, personal data does not belong in it. If the vendor processes data outside the European Union, the GDPR sets extra conditions for that transfer, and the vendor’s documents should say how they are met. This is the point where a short conversation with whoever handles data protection in your company saves trouble later.
What should never be pasted into an AI tool?
Four kinds of content stay out, whatever the plan says. The reason is not that a breach is certain. It is that the damage, if it happens, cannot be undone.
| Kind of data | May it go in? | Why |
|---|---|---|
| Published texts, own marketing drafts, general questions | Yes | Nothing in them is secret |
| Internal procedures and notes without names | Yes, in the approved tool | Low damage, and the terms have been checked |
| Customer or staff personal data | Only with an agreement and a legal basis | The GDPR applies to it |
| Contracts and documents under an NDA | No, unless the other party has agreed | You promised confidentiality to someone else |
| Unpublished figures, prices and plans | No, or only in a tool approved for it | The damage of a leak cannot be reversed |
| Passwords, access keys, bank details | Never | No task needs them, and anyone who holds them holds the account |
A useful habit covers many cases: remove the names. A complaint letter with the customer’s name, address and order number replaced by placeholders can usually be summarised or answered just as well.
How do you make AI use safe in the company?
You choose one tool, check it once and give people a short rule. Forbidding AI tools altogether rarely works: people use private accounts instead, and then nothing has been checked at all.
- Find out what is usedAsk staff which tools they use today and for what. Ask without blame, or you will not get the true list.
- Choose one tool and one planPick a plan with company accounts and check the five points above. Write the answers down.
- Write the ruleOne page: which tool, what may go in, what never goes in, whom to ask.
- Explain itFifteen minutes in a team meeting, with examples from your own work, do more than a document nobody opens.
- Review itLook at the terms and the rule again once a year, and whenever the vendor announces changes.
The rule itself can be this short:
How we do it
When we build AI into a company’s work, we decide first which data the task needs and which it does not, and we go through the five points above with you for the tools involved. Access rights are settled before an assistant is given any internal file. We never share client information.
If an assistant is to answer from internal files, access rights become the central question; see AI assistants for how that is built. Where the documents and backups live is part of the same picture, covered under security and backups.
Questions we hear often
Want this sorted without doing it yourself?
Tell us what you are working on. A person replies within one working day.