Your employees are already using AI on their own. What rules should you give them?

Give people one-page rules that answer three questions: which AI tool they may use, what data they may put into it and who to ask when they’re not sure. The rules should also come with an approved tool on a company account. Without one, people will work around the ban and keep using personal accounts you know nothing about.

How strict your rules need to be depends mainly on the data your company works with. Customers’ personal data, prices and margins, or contracts with confidential terms call for a different approach than mostly public information. It also matters what people already use, how much you’re willing to pay for a company tool, and what your customer contracts and the regulations in your country require of you.

A ban without an alternative just pushes AI into personal accounts

According to a Microsoft survey reported by Euronews, most employees surveyed in the UK use AI tools at work that their employer hasn’t approved. Yet fewer than a third of them are concerned about the privacy of the data they put into those tools. Microsoft warns that this creates a risk of data leaks and cyberattacks.

This isn’t ill will. People have discovered that a tool can speed up writing an email, summarise a long contract or translate a complaint. They use whatever is at hand: a free version in the browser or an app on their phone.

The problem is what they put into the tool. A spreadsheet of customer orders. An email from a customer with their address and phone number. A price list with margins. With a personal account, the company doesn’t know where the data ends up, who has access to it or whether the provider will use it to further develop its model.

A blanket ban usually doesn’t change this. Working faster with AI is a real benefit to people, and they don’t want to give it up. They’ll keep using the tool; they’ll just stop talking about it. And then you know nothing.

The more sensible approach is to give them a permitted route and clear boundaries.

First find out what people already use and what for

Rules written at a desk miss how work is actually done. Before you write them, ask your team. Keep it short, don’t make it an investigation, and promise up front that no one will be punished for being honest.

A few questions are enough:

• Which AI tools do you use at work, paid and free?
• What tasks do you use them for most often?
• Do you sign in with a company or a personal account?
• Have you ever put customer data, prices or internal documents into a tool?
• What work would you like to do with AI officially, if you could?

The answers tell you two things. Where the biggest risk lies today. And what the approved tool must be able to do so people don’t go back to their personal accounts.

If most of the team uses AI to write emails and summarise documents, you don’t need anything complicated. If someone regularly uploads entire customer spreadsheets, you know where to start.

Choose one approved tool with a company account

For a small business’s everyday work, one approved tool is usually enough. More tools mean more terms to read and more accounts someone has to keep track of.

Besides free versions, the major providers also offer business plans. They differ mainly in what happens to your data and who manages the accounts. When choosing, check the terms and settings for:

1. Training on your data. Does the provider use the content you enter to develop its models? For a company tool, you want a clear no, or the option to turn it off for the whole company.
2. User management. Can you add and remove people from one place? When an employee leaves, their access and conversation history must stay under the company’s control.
3. Data storage and deletion. Where and for how long are conversations and uploaded files kept, and can they be deleted?
4. Sign-in. Can you use a company email address and two-factor authentication?
5. Contractual terms for processing personal data. Does the provider offer the agreement or addendum required by the data protection rules that apply to you? Check exactly what you need with a data protection specialist in your country.

A business plan costs money. But compare it with the time people are already saving with AI today, and with what a leak of customer data would cost the company. If you don’t pay for the tool with money, you pay with risk.

Name the approved tool in the rules. The sentence “use only approved tools” without a list helps no one.

Sort data into three groups: yes, after editing and never

For employees, the hardest part isn’t choosing a tool. It’s deciding what they may put into it. A simple classification that everyone can remember helps.

Adjust the boundaries to your industry. A wholesaler will be most sensitive about customer prices and margins. An accounting firm about clients’ records and financial data. An online shop about addresses and order history.

Before pasting any text, people can ask themselves one check question: “Would I mind if a competitor, or the customer it concerns, read this?” If the answer is yes, the text belongs in the second or third group.

Write one-page rules using this template

Replace the square brackets with your own details. You don’t need more than one page.

1. For work, we use [tool name] with a company account. [Who] will set up your access.
2. Don’t put anything company-related into personal accounts or other AI tools. A general question with no company information is fine.
3. We sort data into three groups: yes, after editing and never. The overview is on the other side.
4. Before pasting text, remove names, contact details, account numbers and company names.
5. You are responsible for the result, not the tool. Check facts, figures and quotes before passing them on.
6. Read any text that leaves the company in full before sending it.
7. Want to try a new tool or add-on? Write to [who]. You’ll get an answer by the end of the week.
8. If you’re not sure, ask [who]. If you paste something by mistake, say so straight away. No one is punished for reporting it.
9. We’ll review the rules together again [when].

The rules should also say what you do want AI used for: summarising long documents, first drafts of emails, translations, proofreading, drafting a spreadsheet. People then see that this isn’t a ban but a guide.

Watch out for tools that don’t look like AI. Browser extensions, meeting transcription and smart features in office software also read your content. They fall under the same rule as any other new tool.

Introduce the rules with examples and review them after a while

Few people read a document sent round by email. A short team meeting where you show the rules on real work is better.

Prepare three examples from your own company:

• a customer email, to show exactly what to delete before pasting it in;
• a task that AI is well suited to, where people can see for themselves how it helps;
• a result that looks convincing but contains a mistake, which is why checking is essential.

Appoint one person to answer questions and approve new tools. It doesn’t have to be a technical person. It’s enough that they understand the company’s work and have the time.

After a few weeks, ask the team again. Are you using the approved tool? What’s missing from it? Where are the rules unclear?

If people are going back to personal accounts, it usually means the approved tool can’t do something or is hard to access. Fix the tool or the access, not the people.

Then review the rules roughly twice a year. And also whenever the provider changes its terms or you start working with a new type of data.

Sources

• Most UK employees use AI at work without permission, Microsoft survey finds

Frequently asked questions

Wouldn’t it be better to ban AI for employees altogether?

A complete ban only makes sense where the company works with data that must not go into external tools under any circumstances. Otherwise, people usually find a way around it and you lose visibility. An approved tool with clear boundaries is safer, because you know it’s being used.

What should I do if someone has already put sensitive data into a personal account?

First, calmly establish exactly what was entered, where and when. Ask the person to delete the conversation in the tool, and if passwords or login credentials were involved, change them immediately. If customers’ personal data was involved, consult a data protection specialist about whether you are required to report the incident to anyone.

Is it enough for people to turn off chat history in the free version?

It only helps partially, and it depends on each individual’s discipline. The company has no visibility and no way to enforce the setting. And when someone leaves, the account and its contents leave with them.

Does every employee have to sign the rules?

That depends on how internal policies work in your company and what employment law says in your country, so check with a local specialist. In practice, it helps more than a signature if everyone goes through the rules with examples and knows who to ask. A signature without understanding won’t prevent mistakes.

How can I tell whether people are still using unapproved tools?

You’ll never be completely certain, and monitoring every step would damage trust within the team. A more useful signal is whether people actually use the approved tool and whether they ask questions. If no one asks anything, probably no one is reading the rules.