A free check · no software to buy

A five minute check: is client data safe in your team's AI?

Most firms I speak to have AI in use somewhere, whether or not anyone decided it should be. Whether that matters depends on which tools people are using and what goes into them, and most firms can't answer either question. Here are five checks you can do today.

By Jonathan Gaze · checked October 2026

  1. Find out which AI tools people actually use

    Ask the team, without making it a confession. The usual answers are ChatGPT, Claude or Gemini on a personal account, Copilot through work, or a mixture. You can't set a sensible rule until you know what's in use.

  2. If you're on Microsoft 365, look for the green shield

    Many firms already pay for a business grade AI tool without realising. Sign in to Copilot Chat with your work account and look for a green shield next to New Chat. That shield means Microsoft's enterprise data protection is on: what you type stays under your firm's Microsoft agreement and isn't used to train the AI. For most Microsoft firms this is the safest option, and it's already there.

    Google Workspace firms have the equivalent in Gemini when signed in with a work account.

  3. On any personal account, switch training off

    Personal accounts, free or paid, are consumer products. Several use your chats to improve their models unless you say otherwise. It takes a minute to change:

    • ChatGPTSettings, Data controls, turn off “Improve the model for everyone”.
    • ClaudeSettings, Privacy, turn off “Help improve Claude”.
    • GeminiTurn off Keep Activity in your Gemini Apps Activity settings. Google's own advice is not to enter anything confidential you wouldn't want a human reviewer to see.
    • Copilot on a personal Microsoft accountIn Copilot's privacy settings, turn off model training.

    One catch: rating a reply with a thumbs up or down can send that conversation back to the provider even with training off.

  4. Agree one simple rule

    Switching training off helps, but a personal account still has no agreement with your firm, and nobody at the firm can see what went into it. The same tool can be safe on a work account and not on a personal one, so the rule has to cover both:

    The rule

    No client details in any AI tool unless the firm has approved that tool and you're signed in with your work account.

    If someone wants help with something client related anywhere else, they take the identifying details out first. “A client aged 60 with a pension of around £400,000” works just as well as a name and a policy number.

  5. Think about what leaves when someone leaves

    Chats in a personal account belong to that person. When they move on, everything they ever pasted in goes with them. Work accounts stay with the firm and can be closed. That alone is a good reason to bring AI use onto the firm's own tools.

If it has already happened

The questions people worry about.

Someone has already put client details into a personal AI account. What now?
Don't panic, and don't make it a disciplinary matter, or the next person won't tell you. Find out what went in and which tool and account it was. Delete the chat, and switch training off if it was on. Then write down what happened and pass it to whoever handles data protection at the firm, because they need to decide whether it counts as a data breach.
Does deleting the chat get rid of it?
Partly. Deleting stops it showing in the account, but providers keep deleted chats for a period, often around 30 days and longer in some cases. If training was switched on, what was typed may already have been used. Google keeps chats that a human reviewer has seen for up to three years. So deleting is worth doing, but it isn't the end of the matter, which is why the rule in check 4 matters more than the clean up.
Do we have to report it?
Possibly. Under UK GDPR, sending client personal data somewhere it shouldn't go can be a personal data breach. Every breach should be recorded internally. It only needs reporting to the ICO if it's likely to put people at risk, and then within 72 hours of the firm becoming aware. A first name and a vague question is very different from a full client file with account numbers. Regulated firms may have their own rules on top. This is the point to involve your data protection lead or adviser, not a web page.
Wouldn't it be simpler to ban AI altogether?
It's the most common reaction, and it usually makes things worse. People who find AI useful keep using it, on their phones and personal accounts, where nobody can see. Approving one tool on work accounts and showing people how to use it well gets you both control and the benefit.
Is a paid personal account good enough?
No. Paying for Plus or Pro gets you a better model, not a business agreement. It's still a consumer account between the provider and that one person. The firm has no contract covering it, can't see what's in it, and loses it when the person leaves.
Next

None of this needs to be a big project.

Most firms can sort it in an afternoon, then get on with using AI properly. If you'd like your team walked through it on their own work, with the tools you already have, that's what my sessions do.

Tell me who is in the room See the sessions

Checked against each provider's published terms in October 2026. Providers change these settings and terms from time to time, so check your own accounts. This is practical guidance, not legal advice.