Is AI safe to use at work?

What should I never tell an AI at work?

Anything that isn’t yours to give. At work, most of what you know belongs to someone else: your clients, your colleagues, your employer.

So, never:

• client or customer details: names, accounts, cases, anything that identifies a person;
• a colleague’s personal information: health, pay, performance, a grievance;
• anything under a confidentiality agreement or not yet public: prices, bids, results, plans;
• passwords, access codes or keys, ever;
• anything your employer has told you not to.

A workable test: if you wouldn’t read it out in a busy train carriage, don’t paste it in, unless your employer’s own AI account is set up for it and you have been told it’s allowed.

If your question is about AI at home (does it know my shopping, my bank, my search history?), I drew that answer as a town: Is AI safe to use? Am I safe to use AI?

Can my employer see if I use AI?

Assume yes, if you are on a work device, a work network or a work account. Business AI accounts are run by an administrator inside your organisation, and that is part of the point of them. On some business plans the administrator can search or export what was typed and what came back. Work laptops and networks can be monitored too.

On your own phone, on your own mobile data, with your own account, your employer normally can’t see inside it. But that is where the real exposure sits. If you paste work into a personal account, the work has left the building and nobody at work knows it has gone.

If you don’t know what your employer monitors, ask. It is a fair question, and a good employer will answer it. The more useful question is the one underneath: what am I allowed to use, and for what?

What are the real risks of AI at work?

Not the ones in the films. The risks I see are behaviours, on both sides of the screen.

On the machine’s side: it is confidently wrong. When it reaches the edge of what it knows, it doesn’t stop; it carries on in the same tone as the answers that were right. And it agrees with you, so the draft you were unsure about comes back praised.

On the person’s side: we stop checking. The first twenty answers were right, so the twenty-first goes out unread. We paste in what we shouldn’t, because the box is friendly. And we use it quietly, so nobody else checks either.

The risk lives in the gap between those two: a fluent machine and a busy person who signs. How to catch the machine’s half is in Why AI makes things up.

How do I use AI safely at work?

Five habits, and none of them needs a technical word.

1. Use the account your employer gives you, for work. If there isn’t one, ask before using your own.
2. Leave out what isn’t yours. Write “the client” instead of the client’s name. You get the same help.
3. Treat everything it gives you as a draft until you have checked the facts in it: the figures, the names, the sources.
4. Keep your name on it. If it goes out under your name, you read the whole thing first. “The AI wrote it” is not a defence I would want to rely on.
5. Say when you’ve used it. Not as a confession. So that someone else knows to check.

What should employees know before they start?

Three things, before any of the habits.

It isn’t checking facts. Even when it searches the web first, it writes the most plausible continuation of what you typed. That is usually right and sometimes invented, and it can’t reliably tell you which.

It will be nice to you. That is not a verdict on your work. Agreement is the most likely way to continue a conversation, so agreement is what you’ll get.

You are allowed to say no to it. If it hands you something you can’t check, you don’t have to use it. Using AI well at work isn’t trusting it most. It is knowing exactly where to stop trusting it.

Is AI putting our business at risk?

I don’t know your business, so I can’t tell you. But the risk I would look for first isn’t the tool you bought. It is the use you can’t see: people using personal accounts for work, because there is no work account, or because nobody told them what they could use.

Three questions show you where you stand:

• Which AI accounts are being used for our work, and whose are they?
• What has been pasted into them?
• Who checks what comes out before it reaches a client?

If you can answer all three, you know where you stand. If you can’t, that is where to start. Training is part of it: AI training for leaders and managers.

Do we need an AI policy?

Yes, and a short one. What a staff policy should say, and who is accountable when AI talks to your customers, I have set out in When your AI says the wrong thing to a customer.

Paul Roebuck · Psychotherapist, author and AI coach. I train leaders and their teams on how AI behaves, and on what people do in response.

If you want your people to use AI at work with less risk and less fear, book a 20-minute chat.