Date posted: 19/11/2026

Shadow AI: what it is and how to manage the risks

What to do when people are using AI for work and nobody knows about it.

In brief

  • Shadow AI is the use of AI for work without the organisation’s knowledge or approval.
  • The risks include losing control of work information, using unchecked outputs and allowing AI to take actions without appropriate oversight.
  • A ban alone can push AI use out of sight. Suitable tools and open conversations can help bring it into view.

By Inbal Rodnay

What is shadow AI?

Shadow AI is the AI version of shadow IT.

Team members adopt AI tools because they are useful, sometimes without the organisation’s knowledge or approval.

This may involve a free or personal account on a work computer or personal phone, used outside the organisation’s approved tools and policies.

The work gets done, but the organisation carries the risk without the safeguards, and often has no idea where its information is going.

This article looks at how shadow AI shows up, how common it is, why it’s a risk and how to bring it into the open.

Two ways shadow AI can show up

Asking AI for help. Someone uses an unapproved AI chatbot to draft or analyse something. They might paste in an email or upload a spreadsheet or contract.

Allowing AI to take actions. Someone gives an unapproved AI tool access to files or systems so it can carry out tasks. Depending on its permissions, it might edit documents, move files or update records. Like other automation, it can complete work quickly — and mistakes can affect many files or records just as quickly.

Both can hide inside tools people already use. A free AI note-taker that joins client meetings, a browser extension that reads every page, or an AI feature switched on in an app nobody has assessed can all send work information somewhere you haven’t approved. See: AI meeting assistants: more than a note-taker.

Shadow AI is rarely malicious. People are under pressure to deliver, they know AI can help, and they use the tools they can find.

It’s more common than you think

Jobs and Skills Australia’s 2025 report, Our Gen AI Transition, cites two 2024 surveys that found around 27% of workers, and 21% of white-collar workers, were using generative AI secretly at work.

Why it’s a risk

When people use AI tools the organisation has not approved, on the web or installed on their devices, you can face four risks:

  • You don’t know where the data is going. You may not know how the tool uses or stores work information, or whether it uses that information to train its models. This depends on the tool, account and settings. If someone uses a personal account, the organisation may also be unable to access or manage the work information held there when they leave.
  • You don’t know how the work was done or checked. AI-assisted work may reach clients or internal stakeholders under your organisation’s name without the checks it needs. See: How to check AI’s work.
  • You don’t know what actions it can take. Depending on its permissions, an AI tool may be able to access files, change records or take other actions in your systems. Without assessing it, you may not know what it can do or whether appropriate limits and reviews are in place.
  • You may breach your obligations. Client, customer or employee information entered into an unapproved tool may breach your confidentiality obligations, and sending personal information to an overseas provider can bring specific obligations under privacy law in Australia and New Zealand, including state or territory laws for some public sector agencies. See: Is my data safe to use with this AI tool?

How to deal with it

The first instinct is to block the sites and write a policy that says no AI.

But blocking every AI site is impractical, and a policy that simply says no often pushes AI use out of sight rather than stopping it.

A better approach is to find out what people are using, give them approved tools and set clear rules.

Here are three steps:

  1. Ask. Ask the team what they’re using and for what. A short anonymous survey works well. Keep it to three questions: which AI tools have you used for work in the past month, what did you use them for, and what kind of information did you put in? Make it clear the goal is to find out, not to catch people out. In a very small team, a conversation may work better than a survey. Include any offshore or outsourced teams.
  2. Equip. Give people access to AI tools approved for their work. Start by checking what is available in Microsoft 365 or Google Workspace, if your organisation uses them, and the AI features in software you already use, such as your accounting or practice management platform. Check the terms, data protections and settings for the specific AI features before approving their use. Show people how to use them for everyday tasks. When approved tools meet people’s needs, they have less reason to go elsewhere. See: Five questions to choose your primary AI tool.
  3. Govern. Publish a practical, easy-to-understand AI policy and make sure everyone understands it. Give people a simple way to request a new tool, and answer quickly. If approval takes months, people will stop asking. In the public sector, check first whether a government-wide or agency AI policy already applies. See: What your AI policy has to cover.

How technical controls fit in

Technical controls can help identify unapproved AI use and restrict access. Your IT team or provider can advise what is available.

For example, if your organisation uses Microsoft 365, Microsoft Defender for Cloud Apps includes Cloud Discovery, which analyses network activity to identify cloud services being used, including AI tools.

Microsoft is also previewing a Shadow AI page in the Microsoft 365 admin centre, through its Frontier early-access program. It can detect certain AI agents installed on managed Windows devices, and admins can apply device policies to block some of them. Preview features may change or be withdrawn, and may not be available to every organisation.

Other security tools offer similar discovery features. Products named here are examples, not endorsements. A quick first step: ask your IT team or provider for a list of the third-party apps and AI features staff have connected to work email, calendars and files. It can turn up note-takers and plug-ins nobody knew about. Check Office add-ins too: some AI tools, such as Claude, can be installed as add-ins inside Word, Excel and PowerPoint, and run under that vendor’s terms rather than Microsoft’s.

Coverage depends on the devices and network activity being monitored. Use on personal devices outside that coverage may remain invisible. These tools also require suitable licences and technical setup, so ask your IT team or provider what your current plan supports.

For smaller businesses, ask your IT provider about practical controls to:

  • Require approval before staff install AI apps or browser extensions.
  • Manage staff access to approved tools, for example through single sign-on or an identity and access management service.
  • Monitor or restrict unapproved AI use, including personal AI accounts used on work devices. Monitoring staff activity can raise privacy and workplace obligations, so check these first.

If you don’t have an IT provider, start with the interim rule and the Ask and Equip steps, and ask the vendors of your main software what admin controls they offer.

Check what each control can actually detect or prevent. Use these controls alongside approved tools, staff training and your AI policy.

Your next move

This week, send the three-question survey. Until approved tools are in place, set one interim rule: no client, personal or other confidential information goes into an AI tool the organisation hasn’t approved.

If the organisation hasn’t given people approved AI tools yet, assume they’ve found their own, and make approved tools your priority.

If approved tools are available but people still use others, find out why. Treat what you find as free research: the tools people choose show you where your approved tools, or your training, are falling short. They may need help using the tools you provide, or those tools may lack a capability they need. Provide training where it will help. Where there is a genuine gap, assess the alternatives before approving them. The best of them may earn a place on your approved list.

Sources

Jobs and Skills Australia, Our Gen AI Transition: Implications for work and skills, 2025

Microsoft, Discover AI apps and data

Microsoft, Zero Trust workshop: Enable Shadow AI detection for local agents on endpoints

Microsoft, Cloud app discovery overview

Microsoft, Zero Trust workshop: Block unsanctioned local agents on managed endpoints

Disclaimer
This guide provides general guidance only and is current as at October 2026. Tool names, features and capabilities were correct at that date but may change. The publication therefore focuses on general principles rather than specific feature lists.

This guide is not intended to be comprehensive and is not a substitute for professional, legal or other advice tailored to your circumstances. Use it alongside your organisation’s artificial intelligence policies and your applicable legal, regulatory, professional and ethical obligations. Check current vendor terms, product information and your own obligations before acting on or relying on the information.

While Chartered Accountants Australia and New Zealand (CA ANZ) seeks to ensure that the information is accurate and current, it does not guarantee its accuracy, completeness, currency or suitability for any particular purpose. To the fullest extent permitted by law, CA ANZ disclaims all warranties and liability for any direct, indirect, special or consequential loss or damage arising out of or in connection with any inaccurate statement, opinion, advice or information contained in this publication, any omission from it, or the use of or reliance on this [guide/document] or any material it contains.

© 2026 Chartered Accountants Australia and New Zealand ABN 50 084 642 571

Search related topics