Shadow AI: when employees build apps without permission

Somewhere in your company, there is a good chance an employee has already built an app with AI. Maybe it is a job tracker a crew lead made on a personal account over a weekend. Maybe it is a customer follow-up tool the office manager published and shared by link. It works, people rely on it, and nobody in charge knows it exists.

That is shadow AI. It is not a sign your people are reckless. It is a sign they have work that needs better tools and they found a way to get them. The risk is real, though, because those apps often hold customer and company data with none of the protections you would insist on if you were asked. This guide explains what shadow AI is, how common it has become, what it puts at risk, how to find what already exists, and how to turn it into a safe, approved way to build.

What shadow AI means

Shadow IT is an old term for software employees use without the company's approval, like a personal file sharing account or an unapproved project tool. Shadow AI is the same idea applied to AI. It covers three levels, and the risk rises with each.

• Using AI chat tools on personal accounts. Someone pastes a customer complaint or a price list into a personal chatbot to draft a reply.
• Connecting AI tools to company data. Someone links an AI tool to the company inbox, shared drive or spreadsheet.
• Building apps with AI. Someone uses an AI app builder to create a working tool, puts real records in it, and shares it with coworkers or customers.

This guide focuses on the third level, because a shadow app does not just touch data once. It stores it, shows it to other people, and keeps running after everyone forgets who made it.

How common it is

The numbers from 2026 surveys are consistent: shadow AI is normal, not exceptional.

Two things stand out. First, many small and mid-size businesses are already building with AI, so the question is not whether your team will build but whether you will know about it. Second, people use tools they believe are not allowed. A policy alone does not stop them.

There is also direct evidence of what happens to the apps. In May 2026, security firm RedAccess reported reviewing about 380,000 vibe-coded assets (apps and pages built mostly by describing them to AI). About 5,000 were built for corporate purposes, and 40 percent of those exposed sensitive data.

Why employees build without asking

Understanding the reason makes the fix obvious. People build shadow apps because:

• The work is real and the current tools are bad. Spreadsheets, group texts and email chains break down as a team grows. The person feeling the pain every day is the one who builds the fix.
• Asking takes too long. In MuleSoft's 2022 Connectivity Benchmark, only 37 percent of IT leaders said they completed every project asked of them the year before. In a small company without an IT department, there may be nobody to ask at all.
• The tools make it easy. AI app builders are designed so anyone can go from idea to working app in an afternoon, and many make publishing a link one click.
• It is not clearly forbidden. Most small companies have no written rule about building apps with AI, so people assume it is fine.

None of these reasons are bad. They are the same reasons a company would want its people building better tools. The problem is only where and how they build.

What shadow apps put at risk

The data risk deserves extra attention. Our guide to how AI-built apps expose customer records covers the documented cases. The short version: apps built quickly and published publicly have repeatedly left records readable by anyone, and on many platforms protecting that data is the builder's job. A shadow app's builder usually does not know that.

How to find the shadow apps you already have

The goal is to see everything, not to catch anyone. Make that clear from the first message, or people will hide the very apps you most need to know about.

1. Announce an amnesty. Tell the team you want to know about every tool they built or use, that nobody is in trouble, and that the goal is to keep the useful ones safe.
2. Ask team leads one question. "What links does your team open every day that are not our official systems?" Write down every answer.
3. Review expenses. Look through expense reports and card statements for AI subscriptions and app builder plans over the past year.
4. Ask about data. For each tool you find, ask what records go into it: customers, employees, prices, payments.
5. Check whether it is public. Open each app's link in a private browser window without signing in. If you see data, that app moves to the top of the list.
6. Find the owner account. Note whose account the app lives in, personal or company.
7. Search for your company. Search your company name together with common app builder terms. Internal tools should never turn up in results.
8. Rank by risk. Public apps with customer data first, then apps with employee data, then everything else.

What to do with what you find

Each app falls into one of three groups.

• Useful and important. Rebuild it on an approved platform, move the data, and retire the old version. Thank the person who built it. They found a real problem.
• Useful but minor. Move it when convenient, or retire it if an approved tool now covers the same job.
• Risky or abandoned. Take it offline, export any data you need, and delete the rest. If customer data was exposed, talk to your legal or compliance advisor.

An example of how this plays out

Here is an example, not a real customer, of how shadow building usually unfolds at a 30-person service company.

The office manager is tired of chasing technicians for job notes by text. One evening she describes a job report app to an AI builder on her personal account, publishes it, and sends the link to the crew. Within a month every technician uses it. Customer names, addresses, gate codes and photos of the inside of customers' homes all go into it.

The owner finds out when a customer asks why their address showed up for someone else. A quick check shows the app opens without a sign-in, a former technician still has the link, and nobody but the office manager can change anything. The app is valuable. It is also a liability.

Handled well, this becomes a win. The owner thanks the office manager, the app is rebuilt on an approved platform inside the company with technician and office roles, the old link is shut down, and the office manager becomes the company's first official builder. The crew keeps a tool they like. The owner gets one they can stand behind.

What to tell your team

The message you send matters as much as the policy. Keep it short and warm. Something like:

"Many of you have found smart ways to use AI to get work done, and some of you have built tools the team now relies on. Thank you. We want to keep the good ones and make sure customer data stays safe. Please tell your manager about any AI tool you use or built for work by the end of the month. Nobody is in trouble. From here on, new tools get built on our approved platform, and we will help anyone who wants to learn."

Why bans do not work, and what does

It is tempting to send a memo banning AI app builders. The surveys above suggest why that rarely holds: people already use tools they believe are not permitted, because the work still needs doing. A ban moves the activity further out of sight.

What works is making the approved path easier than the workaround.

1. Pick one approved platform where every app lives inside the company, people join by invite, roles control who can build, and every change is recorded.
2. Write a one-page policy. What data can go into AI tools, which platform to use for building, and who to tell. Our guide to AI app governance without an IT department has a template.
3. Name the people who build. Give builder access to the people who already showed they will. Your shadow app builders are your best candidates.
4. Make app users free to add. If adding a coworker to a tool costs money, people go back to sharing links from personal accounts.
5. Check in quarterly. Repeat the discovery steps above every three months. It takes an hour once the first round is done.

How FORGE handles it

FORGE turns shadow building into sanctioned building by being the easier path for your team and the safer one for the company.

• Every app lives inside your company. Anyone outside it gets nothing back, not even a hint the app exists. Nothing ends up on the open internet by accident.
• Only the people you invite. Your admin invites each person by email, every invite works once, and there is no public sign-up.
• Owners and admins decide who builds. Builders, admins, owners and app users each see and do exactly what their role allows.
• Tools belong to the company. When someone leaves, removing them ends their access immediately and every tool they built stays with the company.
• A complete record. Every step of every build is recorded with who or what did it.
• Easy for the people who were already building. They describe the work in plain words, a panel of top AI models plans the tool, a dedicated AI crew builds it, and every piece is checked for quality and security before it is saved. Most teams have their first real tool the same day.
• Costs you can see. $49 a seat a month for each person who builds, with 3 builds included. App users are free, so rolling a tool out to the whole team is never a budget question.

Onboarding starts with a workflow workshop where our team works with yours on the work that costs you the most time, which is a natural moment to bring your shadow apps into the open. Talk to us to set one up.

Questions people ask

Sources

1. Pax8: Q3 2026 SMB AI Pulse Report
2. PagerDuty: shadow AI workplace survey 2026
3. Deloitte UK: GenAI Workforce Survey 2026
4. Microsoft and LinkedIn: 2024 Work Trend Index
5. RedAccess: shadow AI builders security report
6. Gallup: AI use at work indicator
7. Salesforce: MuleSoft 2022 Connectivity Benchmark