The short answer
Your business teams are already building software with AI tools, and IT has not seen most of it.
Find the apps, rate each one on three simple factors, then retire, tolerate or adopt it. Give people an approved path so the next app is built in the open.
Shadow IT is not new. In my career running technology for large organisations, the spreadsheet with a macro that quietly ran a business process was always there. What has changed is scale and reach. A sales manager can now build a working web app with a login, a database and a connection to the CRM in an afternoon, using Lovable, Replit, Bolt or Cursor. That app can be on the public internet by evening.
Most of these apps are built by capable people solving real problems. The trouble is that nobody has checked who can see the data, where it is stored or what happens when the person who built it leaves.
How to find them
No single source will show you everything. Use several and compare the results.
SSO and OAuth grants
Review which third-party apps staff have authorised against Microsoft 365 or Google Workspace. AI-built apps often ask for access to mail, calendars or files.
Expense and card data
Search for subscriptions to Lovable, Replit, Bolt, Cursor, v0, Supabase, Firebase, Vercel and Netlify. A paid plan usually means something is in use.
DNS and hosting
Look for subdomains pointing at hosting platforms, and for apps on platform domains that carry your company name.
A short, no-blame survey
Ask each team what they have built and what data it uses. Make it clear the aim is to help, not to punish. You will learn more from this than from any scan.
A simple way to rate the risk
Score each app from 1 to 3 on three factors and multiply the scores. Anything that scores 12 or more needs attention first.
| Factor | 1 (low) | 2 | 3 (high) |
|---|---|---|---|
| Data sensitivity | No personal or financial data | Internal business data | Customer, health or financial data |
| Users | Only the builder | A team relies on it | Customers or partners use it |
| Integrations | Standalone | Reads from core systems | Writes to core systems |
Retire, tolerate or adopt
Every app in the inventory gets one of three decisions, agreed with the team that uses it.
Retire
High risk, little value, or a copy of something IT already supports.
Export the data, tell the users why and switch it off.
Tolerate
Low-risk personal and team tools.
Record them in the inventory and move on. Not everything needs to be governed.
Adopt
People depend on it and it solves a real problem.
Bring it under IT ownership, review the security, add backups and monitoring, and give it a named owner.
Guard rails that do not kill useful work
The goal is to make the safe path the easy path. In practice that means a short list of approved AI coding tools, a one-page checklist before an app goes live, and a light review by IT before an app touches personal or financial data. Require company accounts rather than personal ones, so the work does not walk out the door with the builder.
These apps should also fall inside your existing security baseline. The ASD's Essential Eight controls, such as multi-factor authentication, restricting admin privileges and regular backups, apply to an app built in a weekend just as they do to your ERP.
The privacy question
If your organisation is covered by the Privacy Act 1988, personal information in these apps is your responsibility, however the app was built. If an exposed app leads to a breach likely to cause serious harm, you may need to notify the affected people and the OAIC under the Notifiable Data Breaches scheme. Finding these apps before an attacker does is far cheaper than explaining them afterwards.
Getting help
Our AI App Inventory finds the apps, rates each one and agrees a retire, tolerate or adopt decision with the business. It takes three to five weeks and costs A$15,000 to A$25,000 plus GST. The apps worth keeping can then be hardened through our vibe code rescue service, and the policy side sits with our AI governance work. If a team wants to check its own app first, our fix or rebuild guide has five checks anyone can run.
Common questions
What is shadow AI?
Shadow AI is any use of AI tools that IT has not approved or does not know about. It now includes whole applications that business teams have built with AI coding platforms such as Lovable, Replit, Bolt or Cursor, often connected to real customer or financial data.
Should we ban AI coding tools to stop shadow apps?
Bans rarely work. People keep building, just less visibly. A better approach is to give teams an approved path: approved tools, a short checklist, and a light review before an app touches personal or financial data.
How long does an AI app inventory take?
For a mid-sized organisation, finding the apps, rating them and agreeing what to do with each one usually takes three to five weeks. Most of the time goes on talking to the teams that built them, not on technical scanning.
