Shadow AI Discovery and Control
Shadow AI discovery finds the public AI tools your staff use without approval, records what business data goes into them, and gives each one a decision: approve, restrict or block. You end with a written register of AI tools in use, a named owner for each and recommended controls, not a memo asking people to stop.
OxygenIT is ISO 42001 and ISO 27001 certified, as at 12 May 2026. We are based at 7 Horner Street, Papanui, Christchurch 8053, and work with NZ businesses of roughly 20 to 200 staff.
Updated November 2026. Reviewed by Tim Clarkson
Not ready yet? Check whether you are ready first
What shadow AI discovery covers
Shadow AI is any AI tool used for work without the business approving it: a free chatbot, a browser extension, or an AI feature switched on inside software you already pay for. A shadow AI discovery and control service finds those tools, works out what business data has gone into them and gives each one a decision. For a New Zealand business of roughly 20 to 200 staff, the work has three stages. First, discovery, inside access and privacy boundaries agreed in writing. Second, assessment, recording each tool in an AI register with its risk and a named owner. Third, control: approving safe tools, restricting or blocking risky ones, and writing an AI policy staff can follow. OxygenIT is ISO 42001 certified, the international standard for managing AI, so the register and controls follow a system we are audited on. Read more about the OxygenIT team. The discovery method depends on what each client licenses, using the sign-in, network and device records you already have.
This page covers the service; our article explains what shadow AI is and why it matters.
Who it is for
NZ businesses of roughly 20 to 200 staff that suspect AI tools are in use without approval, and want them found and brought under control without banning the useful ones.
What is included
- Agreed scope and evidence sources
- Identifying AI use
- Assessing data exposure
- Prioritised controls
What you need first
A written agreement on what we may access and where the staff privacy boundaries sit. Discovery does not start until that agreement is in place.
How it starts
A 15-minute call about what you think is in use. We then scope the evidence sources and boundaries with you.
Step 1 of 3
Agree access, evidence and privacy boundaries
Evidence sources we agree with you
- Sign-in and app consent records from your Microsoft 365 tenant, to the extent your licences record them
- Web traffic and device records from the firewall, filtering and device management tools you already license
- AI Discovery Tool for the period of engagement
Access and staff privacy boundaries are agreed with you in writing before discovery starts. The agreement sets which records we look at and who sees the results. The aim is to find tools and data exposure, not to watch individual staff.
Discovery only works inside limits everyone has agreed. In this step we agree with you which evidence sources we use, what access we need and who owns the decisions that follow. You tell staff the review is happening. Nothing is collected until the written agreement is in place.
Step 2 of 3
Identify AI tools and data exposure
We find the AI tools staff are using by reviewing the sign-in, network and device records agreed in Step 1, looking for public AI tools, browser extensions and AI features inside software you already use. This is the discovery and assessment stage. Each tool goes into a usage register with the data it may reach, a risk rating and a named owner. Where a tool could reach client or staff personal information, we say so, because the Privacy Act 2020 applies to that information wherever it ends up. The register only covers what the agreed evidence sources show. A personal phone or personal account outside those sources will not appear, and the register says so.
What the usage register records
- Tool in use
- Data it may reach
- Risk rating
- Owner
Step 3 of 3
Prioritise approved tools and controls
This is the control stage. Each tool in the register gets a decision (approve, restrict or block) from the decision owner you named in Step 1, and we prioritise the controls, starting with the tools that reach the most sensitive data. Where staff need an AI assistant for everyday work, Microsoft Copilot as an approved option gives them an approved alternative to public AI tools. The rules go into an AI policy staff can follow. Fixes beyond AI tool controls, such as permissions or data handling, move to AI security for wider remediation.
What happens after discovery?
After discovery, every tool found has a decision and an owner. You receive the usage register, a risk summary and control recommendations, ordered so the riskiest tools are dealt with first. Fixing what the review uncovers beyond those recommendations sits with our AI security service.
Should you ban AI tools?
Usually not, because bans rarely work. Staff who find a tool useful move to a personal phone or a personal account, and you lose sight of it. We would rather help you approve a safe option and block the risky ones than issue a ban nobody follows. The aim is visibility first, then control.
| Decision | When it applies | What happens next |
|---|---|---|
| Approve | The tool does a job staff need, and it handles business data in a way you can accept | Recorded as approved in the register, with an owner and usage rules in your AI policy |
| Restrict | The tool is useful, but only safe for some data, some uses or some staff | Allowed for named uses or groups only, with limits in the AI policy and settings changed where your systems allow it |
| Block | The tool sends business data somewhere you cannot accept, or an approved tool already does the same job | Blocked where your systems allow it, with staff told why and pointed to the approved option |
Deliverables and discovery limitations
At the end you get a register of the AI tools found, a risk summary, control recommendations and a plain statement of what discovery could not see.
What you receive
- Usage register
- Visibility limitations statement
- Risk summary
- Control recommendations
What discovery cannot see
- Personal devices and personal accounts outside agreed evidence sources
- Data already shared with an AI tool, which discovery cannot recall
- Tools in use after the discovery window
- Remediation beyond recommendations, see AI security
What we need from you
- Written agreement on access and privacy boundaries
- Admin access to agreed evidence sources
- A decision owner for approve, restrict or block
- Staff communication about the review
Discovery shows you what happened. It does not undo data that has already left the business, and if client or staff personal information has gone into an AI tool, the Privacy Act 2020 still applies to it. That is worth knowing before you start, and it is the reason to start now.
Price: from $6,000
How the project runs
Shadow AI discovery and control runs as a fixed-scope project in three stages: discovery, assessment and control. We agree the evidence sources, access and privacy boundaries with you in writing first, and nothing is collected until that agreement is in place.
We agree the project dates and milestones with you before work starts.
Ongoing shadow AI monitoring
Price: from $500 a month for a team of up to ten
Monitoring is an optional ongoing service that starts after the project. You do not need it to receive the usage register, risk summary and control recommendations.
Monitoring keeps the usage register current. New AI tools that appear in the agreed evidence sources go into the register, and each one gets a decision (approve, restrict or block) and a named owner.
How often the register is reviewed, and where your business sits in the published bands, are set out before monitoring starts.
Monitoring cannot recall data already shared with an AI tool.
Frequently asked questions
How do you find the AI tools staff are using?
We review the evidence sources you agree to in writing. Typically that means the sign-in, network and device records your systems already keep, and we look for public AI tools, browser extensions and AI features inside software you already use. What we can see depends on what you license. Personal phones and personal accounts outside those sources stay out of view, and we say so in writing. That gap is one reason the work ends with an AI policy staff can follow.
Will staff know a discovery is happening?
Yes, they should, and we ask you to tell them before it starts. We agree access and staff privacy boundaries with you in writing first, including which records we look at and who sees the results. The aim is to find tools and data exposure, not to catch people out. If you want accountability for AI set out across the business, ISO 42001 AI governance is the next layer.
Should we ban AI tools or approve some?
Approve some, restrict others and block the risky ones. An outright ban rarely works: staff who find a tool useful move to a personal phone or a personal account, and you lose sight of it. We would rather help you approve a safe option than issue a ban nobody follows. If you already run Microsoft 365, our page on Microsoft Copilot for NZ SMBs covers one option you can approve.
What happens after discovery?
After discovery, each AI tool found gets a decision (approve, restrict or block) and a named owner. You receive a usage register, a risk summary and control recommendations, with the riskiest tools first. Fixing wider issues the review uncovers, such as permissions or data handling, is the work of AI security for NZ businesses. Ongoing shadow AI monitoring after the project is optional.
Can discovery find everything?
No. Discovery only sees what the agreed evidence sources record. It cannot see personal devices or personal accounts outside those sources, or tools staff start using after the discovery window. Data already shared with an AI tool cannot be recalled. You get a written statement of these limits with the register, so nobody reads it as complete. Reducing exposure from here is the job of controls, policy and AI security.
What is shadow AI?
Shadow AI is any AI tool used for work without the business approving it, such as a free chatbot, a browser extension or an AI feature switched on inside software you already pay for. It matters because business data can go into those tools without anyone deciding it should. This page covers finding and controlling it. For the background, read about shadow AI in your business.
Book a 15-minute call
Book a 15-minute call to talk through which AI tools you think staff are using, or call our sales line on 0800 242 206.
Or call 0800 242 206
