Shadow AI: The AI Already Running in Your Business

Most Australian business owners could not write down every AI tool their team used this week. They run a tight ship. The tools arrived quietly, inside software the business already pays for, on phones nobody manages, and through people who were only trying to get their work done faster.

That is shadow AI. It is the AI running in your business that nobody approved, and in most Australian businesses there is more of it than the owner expects. The useful question is not how it got in. It is what it is telling you.

What shadow AI actually looks like

Put aside the picture of someone doing something reckless. In a business of ten to eighty people, shadow AI is almost always three ordinary things.

Someone is paying for it themselves. A team member has ChatGPT or Claude on their own card, because it makes Tuesday easier and asking felt like a production. Your documents go into it. On a personal plan, you have far less say over what happens to what you paste in than you do on a business one.

Your existing software switched it on. This is the big one, and it is not an employee decision at all. Your practice management system, your CRM, your accounting package, Microsoft 365. Someone with admin rights ticked a box, or the vendor turned a feature on by default in an update nobody read. The AI is now inside a system you already trusted, with access to everything that system can see.

Something still has access from a trial two years ago. An AI tool was connected to the mailbox or the file drive during a free trial. The trial ended. The connection did not. Nobody remembers granting it, which is exactly why nobody has revoked it.

None of those three is misconduct. Two of them are not staff actions at all.

Why a ban does not hold

The instinct is to shut it down. Block the sites, send the email, put it in the handbook.

It moves the behaviour without removing it. Block a tool on the office network and the phone in someone's pocket has a camera and a personal subscription. Now the same document is going to the same place, and you have lost the one thing you had, which was the chance to know about it.

Think about the worn track across a lawn. There is a paved path, and twenty metres away there is a strip of dirt where everyone actually walks, because it goes where they are going and the path does not. You can put up a sign. You can put up a little fence. The track stays, because the track is shorter. The only thing that ever moves people back onto the path is moving the path.

The Permission Gap

There is one question that tells you more about how much shadow AI you have than any audit will.

If someone here wanted to try a new tool, who do they ask, and how long does it take to get an answer?

Take that answer, and hold it next to how long a person will wait when they have something due Thursday. The distance between those two numbers is your Permission Gap. Shadow AI grows in it, at exactly the rate you would expect.

If the approved way to get a tool takes three weeks and a partners' meeting, and the unapproved way takes ninety seconds and a credit card, you do not have a discipline problem. You have an arithmetic problem, and the arithmetic will win every time.

This is the part most businesses get backwards. They respond to shadow AI by tightening the rule, which widens the gap, which produces more shadow AI. The fix is almost never a stricter policy. It is a faster yes: someone who can approve a low-risk tool this week, a short written list of what is already fine to use, and a real answer for the rest inside a few days.

Where the Australian rules actually sit

Worth being precise here, because there is a lot of loose talk about it.

In October 2025 the National AI Centre published Guidance for AI Adoption, which sets out six essential practices for organisations using AI. It replaced the earlier ten voluntary guardrails, and it is voluntary guidance.

The fourth of those practices is share essential information, and part of it is keeping a record of where AI is used in your organisation. That is the whole reason shadow AI matters beyond the security question. The guidance asks for a record, and most businesses cannot produce one, because they genuinely do not know.

Then on 15 July 2026 the Prime Minister announced a set of Australian Standards for AI and an Office of AI inside the Department of Prime Minister and Cabinet. National Cabinet is considering the approach this month, and legislation is expected to be introduced early next year, with commencement staged after that.

So, plainly: nothing in that framework binds your business today, and the detail is not settled. Anyone telling you that you are already non-compliant with an AI law is selling you something. The reason to know what AI is running in your business is that it is your business, your client information and your professional obligations. The regulation, when it lands, will only ask you to write down what you should already be able to answer.

Your shadow AI is a map

Here is the part worth sitting with.

Every unapproved tool in your business is a person telling you where the work hurts. Nobody goes looking for a workaround on a task that is already easy. They went looking because something was slow, or repetitive, or landing on them at 6pm.

That makes your shadow AI the cheapest evidence you will ever get about where AI is worth building properly. It is a record of what your people reached for when nobody was helping them, and they paid for the research themselves.

If six people are quietly using AI to draft the same kind of letter, you have just been handed your first build, with the demand already proven and the objections already gone. If one person has found a tool that saves them four hours a week, you have found something better than a tool. You have found the person who makes the next thing land. Every business that has genuinely got AI working has one of those, and they are almost never the most senior person in the room.

Go looking with that posture and people tell you the truth. Go looking like an auditor and the honest answers stop, permanently.

How to find yours in an afternoon

There is an entire industry selling network monitoring and endpoint software for this. It is built for organisations with tens of thousands of staff, and a thirty-person business does not need any of it.

Four places, in this order, because this order finds the most in the least time.

1. Your Microsoft 365 or Google admin. Look at the third-party applications that have been granted access to your mail, files and calendar. This is where the forgotten trial connections live.
2. Your own browser. Open the extensions page and look at what is installed, then check which account you are actually signed into. Personal or work matters more than people think.
3. Your card statements. Search your accounting package for the obvious vendor names. Remember that the subscriptions you find are the ones on the business card, and the ones you will not find are the point.
4. The settings of the software you already pay for. Sit with whoever administers each system and look at what AI features are switched on.

Then have the conversation, which finds more than the four places combined. Two questions do most of the work. When someone here gets stuck on a first draft, what do they actually do? And has anyone found a tool that genuinely saves them time?

Ask what people are banned from using last. Ask it first and every earlier answer gets edited.

What to do with what you find

Put it in one table, and give every row one of three answers, and make them verbs. Telling a business owner that something is "medium risk" gives them nothing to do on Monday.

• Keep and support. It is doing real work and the risk is acceptable. Make it official, put it on the business account, write down what may and may not go into it, and make it easier to use than the alternative.
• Move it. The job is legitimate, the tool is the wrong home for it. Name where it should go instead, in the same breath.
• Stop it. Rare, and reserve it for genuine exposure. Client-identifying material going into a personal account. Access granted to something nobody can now identify. When you stop something, say what replaces it, or you have just built a new track across the lawn.

That table is the record the guidance asks for. It is also, more usefully, your build list.

Your next move

You do not need a policy, a committee or a platform to start. You need to know what is actually running, who is running it, and which one of those things is worth doing properly.

That is part of what an AI Scan does. We look at how your business actually runs, find the AI that is already in it, and tell you honestly which of it to keep, which to move, and where building something properly would pay back first. You keep the register and the map.

If you want to know what is running in your business, book an AI Scan, or see how we help.

Frequently asked questions

What is shadow AI?
Shadow AI is any AI being used in a business that has not been approved or is not known about. In Australian businesses it is usually one of three things: a staff member paying for a tool on their own card, an AI feature switched on inside software the business already owns, or an AI tool that still has access to the mailbox or file drive from a trial that ended long ago.

Is shadow AI actually a problem?
It carries real risk, mostly around what business and client information ends up in a personal account you have no control over, and around access nobody remembers granting. It is also useful information. Every unapproved tool is a person showing you where the work is slow, so the sensible response is to find it and decide what to do with it.

Should we just block AI tools at work?
Blocking moves the behaviour. It does not remove it. Block a tool on the office network and the same document goes through a personal phone instead, and now you cannot see it at all. The thing that works is making the approved option genuinely easier and faster to get than the unapproved one.

Does Australian law require us to keep a record of our AI use?
Not today. The National AI Centre's Guidance for AI Adoption, published in October 2025, asks organisations to share essential information about where AI is used, and it is voluntary guidance. The Government announced a set of Australian Standards for AI on 15 July 2026, with legislation expected to be introduced early in 2027 and commencement staged after that. Your existing obligations around privacy and client confidentiality already apply to AI use regardless.

How do we find the shadow AI in our business?
Start with four places: the third-party apps with access in your Microsoft 365 or Google admin, the extensions in your browser, AI subscriptions on your card statements, and the AI settings inside the software you already pay for. Then talk to your team about what they actually do when they get stuck. The conversation usually finds more than the audit.