WHITEBOXX / GUIDE
Shadow AI: what to do about
AI-built tools nobody owns.
Shadow AI is any AI tool or AI-built software that people use at work without the IT or security team knowing. AI coding tools make it easy to build internal tools quickly. Many of those tools have no owner, no tests and no record of what they do.
What is shadow AI?
Shadow AI covers two things. The first is staff using AI services, such as chat assistants, with company data the organisation has not approved. The second is newer: internal tools and scripts built quickly with AI coding assistants and put into use without review.
The second kind is harder to find. It looks like ordinary internal software, but nobody may understand the code, and the person who built it may have left.
Why it is a growing problem.
In Retool’s 2026 State of AI Governance report, which surveyed 307 technical and security leaders, 93% said they were concerned about AI-built internal tools running in production, and 92% said their governance was not strong. Retool sells tools for building internal software, so treat its figures as a signal rather than a measurement.
Governance guides point to the same root cause: software with no named owner. A tool nobody owns is not patched, not tested when data changes and not retired when it stops being useful.
TAKE STOCK
How to take stock of AI-built tools.
- 01
List the tools
Ask each team which AI tools and AI-built apps they use for work, including spreadsheets with scripts.
- 02
Find the data each one touches
Mark the tools that read customer, financial or personal data.
- 03
Name an owner
Every tool that stays needs a person responsible for it.
- 04
Check what it does
Someone other than the builder should be able to explain the logic and test the result.
- 05
Decide its future
Retire it, adopt it with an owner and tests, or rebuild it properly.
DECISION TABLE
Retire, adopt or rebuild?
| Situation | Decision |
|---|---|
| Few people use it and it touches no sensitive data | Retire it, or leave it with the team and a named owner |
| It works, people rely on it, and someone can maintain it | Adopt it, with an owner, tests and access controls |
| People rely on it, but nobody can explain or maintain it | Rebuild it on a supported platform |
| It produces results that need to be checked later | Rebuild it with a record of sources, steps and outputs |
Where a platform helps.
Rebuilding on a shared platform gives each tool the same data access, access controls and run records, so your team does not build them again for every tool. Evidence OS records the inputs, steps and outputs of each workflow run. Our engineers can rebuild a tool your team depends on as an extension of the platform.
See how our engineers work