Is it safe to use ChatGPT and Copilot at work?

ChatGPT, Microsoft Copilot, Claude and Google Gemini can all be used safely at work, as long as staff use business accounts rather than personal ones, keep sensitive information out of tools you haven't approved, and follow a short written AI policy. The biggest risk is client, staff or confidential information being pasted into free personal accounts, where conversations can be used to train the provider's models. With Copilot and Gemini in Google Workspace, the main risk is internal: they can surface any file a user already has access to.

How many businesses manage the risk?

In the government's Cyber Security Breaches Survey 2025/26, 31% of businesses were using AI, adopting it or actively considering it, but only 24% of those had cyber security practices in place to manage the risks from using it.

And that's only the AI the business knows about. Many staff use free tools on their own initiative, which is where most of the risk sits.

The real risks

• Data leaving your control. The NCSC's guidance on large language models points out that what you type is visible to the provider and stored, and will almost certainly be used to develop the service at some point. Its advice: don't include sensitive information in queries to public AI tools, or anything that would cause problems if it were made public.
• Personal data. Client records, staff details and anything else covered by UK GDPR still need protecting when they go into an AI tool. Using a tool without suitable terms can put you in breach.
• Oversharing inside your business. Copilot and Gemini in Google Workspace work with the files and emails each person can already reach. If permissions are too loose, such as salary spreadsheets shared with "everyone", AI makes them much easier to find.
• Confident wrong answers. AI tools can produce plausible but incorrect information. Anything going to a client, into a contract or into your accounts needs a human check.
• Unapproved tools. Browser extensions, AI note-takers that join meetings and new apps appear every week, each with its own data terms.

Personal accounts vs business accounts

The same product can handle your data very differently depending on the account:

Sources: OpenAI's help centre, Microsoft's enterprise data protection documentation, Anthropic's model training page, Google's Workspace with Gemini FAQ and the Gemini Apps privacy notice, at the time of writing. Providers change their terms, so check before you rely on them.

Not being used for training doesn't mean nobody else can ever see your data. Business accounts still store it, under the provider's business terms. That's why the policy below matters as much as the account type.

What about Claude and Gemini?

The same rules apply, with a couple of points specific to each:

• Claude. Anthropic's Team and Enterprise plans give you a business account with an admin console, so you can control who has access. Claude can also be connected to other services, such as Google Drive, Gmail or Microsoft 365. A connected Claude can read whatever that person can reach, so decide which connectors are allowed before staff switch them on.
• Gemini and Google Workspace. If your business runs on Google Workspace, Gemini is built into Gmail, Docs, Drive and Meet, and covered by your Workspace terms when staff use their work account. Your admin can turn Gemini on or off for different groups of staff. The risk to watch is staff using the Gemini app on a personal Gmail account for work, where Google's consumer terms apply instead.

Settings worth checking first

1. Give staff an approved tool with a work account. If your business uses Microsoft 365, make sure staff sign in to Copilot Chat with their work account. On Google Workspace, the same goes for Gemini. That way your business terms apply.
2. Tidy up file permissions before Copilot or Gemini. Review who can see what in SharePoint, OneDrive and Teams, or in Google Drive and shared drives, especially files shared with everyone or by "anyone with the link".
3. Turn on multi-factor authentication for every account that can use AI tools. A hijacked account with Copilot can search your files in seconds.
4. Control unapproved AI apps. Web filtering lets you allow the AI tools you've approved, block the rest, and control what data can be copied or uploaded to them.
5. If personal accounts are allowed at all, ask staff to switch off model training in their settings, and never to use them for client or personal data.

What to put in an AI policy

A one-page policy is enough to start. It should cover:

• Which tools are approved, and which account to use for each.
• What must never go in: client personal data, staff records, passwords, financial details, confidential contracts, and anything covered by a non-disclosure agreement, unless it's in an approved tool cleared for that use.
• Checking the output: a person reviews anything AI-generated before it goes to a client or is relied on.
• Meeting recorders and note-takers: whether they're allowed, and telling attendees when one is used.
• New tools: who to ask before trying one, and how quickly you'll answer.
• Reporting mistakes: what to do if someone has pasted something they shouldn't, without fear of blame.

Review it every six months or so: the tools change quickly.

Where we can help

Our AI adoption consultancy helps you choose the right tools, set up business accounts securely, write your AI policy and train staff on real tasks. If you'd like a check of your Microsoft 365 or Google Workspace permissions before rolling out Copilot, talk to us.

< Back to Guides