Shadow AI: The Unapproved Tools Putting Your Business at Risk

You almost certainly have not approved every AI tool your team is using. That is not a criticism, it is just how AI adoption has happened in most Australian businesses: bottom up, one free account at a time, with nobody in the loop deciding whether it was a good idea.

This is shadow AI, and it has quietly become one of the biggest gaps between how businesses think they operate and how they actually operate. As far back as 2023, ISACA’s survey of Australian and New Zealand professionals found that around 63% of employees were already using AI at work, while only 11% of their organisations had a formal policy covering it. A 2025 Josys survey of Australian tech decision-makers found the gap has not closed: more than one in three professionals regularly upload sensitive company data into AI tools without oversight, and 70% of organisations reported only moderate to no visibility into which AI tools their staff actually use.

None of this requires anyone to be reckless. A bookkeeper pasting a supplier invoice into ChatGPT to draft a follow-up email is trying to save time, not create a compliance problem. But multiplied across a team, over months, without visibility or rules, that is exactly what shadow AI becomes.

The Numbers

What Counts as Shadow AI

Shadow AI is any AI tool being used for work that the business has not vetted, approved, or accounted for. It does not need to be sophisticated. The most common forms are the least dramatic:

A salesperson using a free transcription app to record client calls, without checking where the recording is stored. A marketing coordinator pasting a customer list into an AI tool to draft personalised emails. An HR contractor running resumes through an AI screening tool that was never assessed for bias. A finance assistant uploading a spreadsheet of supplier pricing into an AI tool to reformat it. Each of these looks like ordinary productivity. Each of them also moves business or customer data outside the business’s control, into a system with its own retention policy, its own training practices, and its own jurisdiction.

The distinction that matters is not whether AI is being used. It is whether anyone in the business knows it is happening and has made a deliberate decision that it is acceptable.

Why It Keeps Growing, Even Where There Is a Policy

Shadow AI persists for the same reason shadow IT always has: the unapproved option is usually faster than the approved one. If the sanctioned tool requires a request form, a licence, and a two-week wait, and the free alternative is one browser tab away, most employees will choose speed. This is not defiance. It is a rational response to friction.

It also grows because AI tools now arrive embedded inside software your team already uses. A CRM adds an AI summary feature. A scheduling tool adds an AI assistant. Nobody signs up for a new account, so nobody thinks to flag it, but the data is still flowing to a third party. A policy written in 2024 that lists five named AI tools has already fallen behind the AI features quietly switched on inside your existing stack.

The Privacy Act Angle

Shadow AI is not just an efficiency or IT hygiene issue in Australia. It is a growing compliance exposure. The Privacy Act’s automated decision-making transparency obligations take effect on 10 December 2026, requiring disclosure when a computer program uses personal information to make or influence a decision that significantly affects someone. If a shadow AI tool is quietly screening job applicants, prioritising customer complaints, or scoring leads, and nobody in the business knows it exists, you cannot disclose what you do not know is happening.

The OAIC has consulted on transparency guidance for this requirement, with the consultation closing in June 2026 and formal guidance expected later in the year. But the underlying obligation does not wait for that guidance to land: the OAIC has already made clear that existing privacy principles apply to AI use today, sanctioned or not. If personal information is going into a tool you have not assessed, you are exposed under APP 8 (cross-border disclosure) and APP 11 (security) regardless of whether anyone approved the tool.

This is also why an audit needs to happen before December, not after. Finding out what shadow AI exists in your business after a complaint or a breach is a materially worse position than finding out now.

Small Businesses Are More Exposed, Not Less

It is tempting to assume shadow AI is a large-enterprise problem, with thousands of employees and dozens of departments each finding their own tools. The opposite tends to be true for risk. Industry research on business AI readiness has found that only around 30% of companies with fewer than 250 employees feel equipped to assess AI risk, compared with 42% of larger organisations. Larger businesses at least have a security or compliance function that might notice. Most Australian SMEs do not, which means shadow AI can operate for a long time before anyone in the business has a reason to look for it.

How to Get Visibility: A 5-Step Audit

1. Ask your team directly. A short, non-punitive survey asking what AI tools people use for work and what data goes into them will surface most shadow AI, because most of it is not hidden on purpose. Frame it as a fact-finding exercise, not a disciplinary one, or people will simply stop answering honestly.

2. Check the paper trail. Review expense claims and card statements for AI subscription charges, and check browser extensions installed on company devices. Shadow AI usually leaves a billing trail even when it does not leave a policy trail.

3. Audit your existing software for embedded AI features. Go through your CRM, helpdesk, accounting platform, and scheduling tools and check what AI features have been switched on, by default or otherwise. This is the fastest-growing source of shadow AI and the easiest to miss.

4. Classify what you find. Not every unapproved tool is a serious risk. Sort what turns up into three tiers: fine as-is, fine with a data restriction, and needs to stop immediately. A tool used only for internal brainstorming is a different risk to one processing customer records.

5. Replace, don’t just ban. If you remove a shadow tool without offering an approved equivalent that is genuinely as fast, it will come back under a different name within weeks. Pair every restriction with a sanctioned alternative, then put it in writing with an AI usage policy your team can actually follow.

The Bottom Line

Shadow AI is not a sign that your team is reckless. It is a sign that AI adoption has outpaced most businesses’ ability to track it, and that the approved path has usually been slower than the unapproved one. The fix is not a memo telling people to stop. It is an honest audit of what is already happening, a fast and genuinely usable set of approved tools, and a policy that keeps up with the AI features quietly arriving inside the software you already pay for.

Treat it as the entry point to a wider AI governance framework, not a one-off clean-up. Shadow AI is what happens in the gap where governance should be.

Frequently Asked Questions

Related Articles