Shadow AI: what it is, why it happens and how to fix it.
Your people are already using AI. The question is whether you can see it.
The short answer
Shadow AI is staff using AI tools, such as ChatGPT, Claude or Gemini on personal accounts, that the business has not approved or cannot see.
It happens when people want to work faster and have no approved option.
The risk is not the AI itself but customer, staff and commercial information going into tools nobody is managing.
Key facts.
- Only 12% of New Zealand employees use only company-provided AI tools, against 20% globally: the lowest of any market in Qualtrics' 2026 Employee Experience Trends study (934 NZ respondents).IT Brief NZ, reporting Qualtrics, 3 March 2026 · source
- 68% of New Zealand organisations with 100 or more staff use general-purpose assistants such as ChatGPT, Gemini and Claude.Datacom State of AI Index 2026, 207 senior leaders, fieldwork July to August 2026 · source
- 75% of knowledge workers used AI at work, and 78% of those AI users brought their own tools.Microsoft and LinkedIn Work Trend Index, 31,000 people in 31 countries, May 2024 · source
- In a UK and US survey by a security vendor, 49% of employees used AI tools their employer had not sanctioned, and 51% had connected AI tools to other work systems without IT approval or oversight.BlackFog (vendor research by Sapio Research), 2,000 employees, January 2026 · source
- The Privacy Commissioner's starting point is that the Privacy Act applies to everyone using AI tools in New Zealand, and its advice is that if in doubt, you do not use AI tools to handle personal information.Office of the Privacy Commissioner, September 2023 · source
What is shadow AI?
Shadow AI is the AI version of shadow IT: technology people use for work without the business knowing, approving or managing it.
It is usually well meant. Someone pastes a customer complaint into ChatGPT to draft a reply, uploads a spreadsheet to summarise it, or connects an AI assistant to their work email to save an hour. Most of it involves generative AI, the kind that writes, summarises and answers questions. New to the terms? Start with our plain English guide to what AI is and how New Zealand businesses use it.
It covers more than chat
Shadow AI also includes:
- Personal accounts on tools the business does use, outside its controls.
- AI features switched on inside approved software, such as note-takers joining meetings, without anyone deciding they should be.
- Browser extensions and plug-ins that read whatever is on screen.
- AI agents and connectors given access to email, files or systems. Our guide to MCP and AI connectors explains why these matter more than chat, and our guide to AI agents covers what they can do on their own.
Why does shadow AI happen?
Because people are trying to do a good job.
Qualtrics put the New Zealand gap down to organisations not providing approved alternatives, as reported by IT Brief NZ, and that matches what we see in our own work with New Zealand organisations. Shadow AI grows where three things are true:
- The tools are free and easy. Anyone can open ChatGPT in a browser tab in a minute.
- There is no approved option, or the approved one is harder to use than the free one.
- There are no clear rules, so people guess, and most guess that it is fine.
Banning AI rarely works. It pushes use onto phones and personal laptops, where you see even less. Overseas vendor research points the same way: in BlackFog's UK and US survey, 63% of employees said it was acceptable to use AI tools without IT oversight if no approved option was provided. That is a security vendor's own survey, not New Zealand data, but the pattern is familiar.
What are the risks of shadow AI?
Five risks, and the last is the one most leaders miss.
| Risk | What it looks like | Why it matters |
|---|---|---|
| Privacy | Customer or staff details pasted into a free tool | The Privacy Act 2020 still applies, and you may not know where the data went. See data privacy. |
| Confidentiality | Pricing, contracts or board papers summarised in a personal account | Free plans may use prompts to improve models, depending on the provider and settings. |
| Security | An extension or agent given access to email or files | A connected tool can read and act on far more than one pasted paragraph. See prompt injection. |
| Quality | AI output sent to a client unchecked | Nobody reviewed it, and nobody knows AI wrote it. AI can hallucinate with confidence. |
| Lost value | Your best people find great uses and keep them to themselves | The wins live with whoever is using AI that day, not the business. |
The last one is the risk most leaders miss. Shadow AI is not only exposure. It is also capability the business cannot see, share or build on.
Shadow AI in New Zealand.
New Zealand stands out: Kiwi staff want to use AI, and many workplaces have not yet given them a safe way to.
In Qualtrics' 2026 research, New Zealand had the lowest share of employees using only company-provided AI tools of any market surveyed: 12%, against 20% globally. Read that as demand, not bad behaviour. Use is already widespread: Datacom's State of AI Index 2026 found 68% of New Zealand organisations with 100 or more staff use general-purpose assistants such as ChatGPT, Gemini and Claude.
What the Privacy Commissioner expects
The Privacy Commissioner's guidance on AI and the information privacy principles is the clearest New Zealand benchmark for what good looks like. In summary:
- The Privacy Act applies to everyone using AI tools in New Zealand.
- Do a privacy impact assessment before you start using an AI tool, and keep it updated.
- The information privacy principles apply at every stage, from what goes into a tool to acting on what comes out, and personal information, including prompts, must be kept secure.
- If in doubt, do not use AI tools to handle personal information.
Shadow AI makes every one of those hard, because nobody has assessed the tool and nobody knows what went into it. Our guide to AI and the Privacy Act 2020 goes through each principle in plain English.
Not legal advice. This is general information as at the date above. For decisions about your organisation, talk to a lawyer.
Our view.
Experimenting is the riskiest stage, not the safest. Leaders often treat a few people quietly using ChatGPT as cautious and low risk. We see it the other way. In the five stages of AI maturity (Aware, Experimenting, Adopting, Adapting, Frontier), Experimenting is where risk is live, invisible and growing: company context flowing into tools nobody monitors, with no policy and no visibility.
The fix is not a ban. It is governance, and governance speeds things up. Once people know what is allowed, managers can say yes to AI in real work instead of leaving it in the shadows.
And a policy document is a start, not a control. Real controls are technical: approved accounts, data settings and access you can switch off.
Where should you start?
Fixing shadow AI takes five steps, and the first is finding out what is already happening.
- Find out what is being used. Start with the free NZ AI Benchmark: 40 questions, takes about 20 minutes, and gives you a baseline. Then ask your people, without blame. An anonymous two-minute survey usually surfaces more than an IT audit.
- Give people a better approved option. A business plan of ChatGPT, Claude or Microsoft Copilot, with data controls switched on, beats a ban every time. Compare plans and data handling in our AI tools comparison for NZ businesses.
- Write short, clear rules. What is in, what is out, and which information never goes into AI. Start with the master prompt for a basic AI policy, then use the free AI Champions Committee Resource Pack for the full policy, register and charter. Our guide on how to write an AI policy has more.
- Set the technical controls. Single sign-on, admin control over connectors and extensions, and data retention settings. Rules people can ignore are not controls.
- Turn the shadow users into champions. The people already using AI are your fastest route to good practice. Give them a role, a channel to share what works, and training: a Lunch n Learn, Claude at Work or the Microsoft Copilot course.
Once the basics are in place, the Organisational Brain Toolkit gives the approved tools your company context, so they are better than the free ones, not just safer. Want to talk it through? Book a 20 minute AI Chat with us.
Frequently asked questions.
What is shadow AI?
Shadow AI is staff using AI tools at work that the business has not approved or cannot see, such as ChatGPT on a personal account or an AI note-taker nobody signed off. It is usually well meant: people want to work faster and have no approved option. The risk is the information that goes into tools nobody is managing.
Is shadow AI illegal in New Zealand?
Using an AI tool is not illegal in itself. The risk is what goes into it. If personal information is involved, the Privacy Act 2020 applies, and the Privacy Commissioner recommends a privacy impact assessment before you use an AI tool and, if in doubt, not using AI tools to handle personal information. This is general information, not legal advice.
Should we ban ChatGPT at work?
Bans rarely work. They push use onto personal phones and laptops, where you have even less visibility. A better approach is an approved business tool with data controls, short clear rules, and training, so the safe option is also the easiest one.
How do I find out if my staff are using AI?
Ask them, anonymously and without blame. A short survey on which tools people use, for what, and with what kind of information usually tells you more than IT logs. Treat the answers as a map of demand, not a list of offenders.
What is the difference between shadow AI and shadow IT?
Shadow IT is any technology used without approval. Shadow AI is the AI part of it, with an extra risk: AI tools take in information to do their work, and agents and connectors can act on email, files and systems, not just store data.
Do free AI tools use our data for training?
It depends on the provider, the plan and the settings, and policies change. Business and enterprise plans generally offer stronger data controls than free personal accounts. Check each provider's current terms before approving a tool, and read our explainer on trust concerns with popular AI assistants.
Sources.
- NZ workers keen on AI but turn to risky shadow tools, IT Brief NZ, reporting the Qualtrics 2026 Employee Experience Trends Report (3 March 2026; accessed 21 October 2026)
- State of AI Index 2026, Datacom (2026; accessed 19 October 2026)
- Microsoft and LinkedIn release the 2024 Work Trend Index on the state of AI at work, Microsoft (8 May 2024; accessed 21 October 2026)
- BlackFog research: shadow AI threat grows, BlackFog (UK and US vendor research) (27 January 2026; accessed 21 October 2026)
- Artificial intelligence and the information privacy principles, Office of the Privacy Commissioner (September 2023; accessed 21 October 2026)
- Privacy Act 2020, Parliamentary Counsel Office (latest version; accessed 21 October 2026)
Related topics
Keep going.
Your next step
Bring shadow AI into the open.
The free AI Champions Committee Resource Pack gives you the policy, the register and the charter to govern AI use, with editable templates for each.
Changelog: 21 October 2026, first published.