Resolon Start a conversation

InsightsAI governance

Published 7 min read

Shadow AI: find it, then govern it, don’t ban it

A ban pushes AI use onto personal accounts, out of your sight. Count what exists, keep what helps, and set the line in writing.

Chad Neptune, CEO and Chief AI Architect

Contents

You almost certainly have shadow AI in your company right now. The real question is whether you know where.

It is already there

A salesperson has a customer spreadsheet and a deadline. They paste it into a free chatbot on their personal account, get a summary in a minute, and move on.

Nobody told them to. Nobody told them not to. And nobody else knows it happened.

That is not a story about a bad employee. It is a person trying to do the job well with the tools they could find. Multiply that by every team and you have shadow AI.

That scene is an illustration, but it is pretty much what I see happening in business after business. In highly regulated industries, I have seen people paste Social Security and driver’s license numbers into these tools, with no guardrails. A model will often reject the request, but that does not change the fact that the personal information went in up front. I have watched companies in those industries lose their cyber insurance after an audit turned this up. It is a liability issue and a security issue at once.

For leaders Assume it exists before you go looking, and make the looking safe for the people who did it.

Two colleagues at a white table with printed charts, a phone, a tablet and a laptop between them

Why a ban does not work

The first instinct is to ban it. I understand the instinct. It feels like control.

But a ban does not stop the work. It moves it. The same people, with the same deadlines, use the same tools on their phones and personal accounts, and now you cannot see any of it.

It also costs you the people who understand your processes best. They are often the ones experimenting, and the ones who know where the bottlenecks are.

A stronger answer is to block the personal tools and provide an approved one. Where the approved tool does the job, that is the Move decision below. Blocks reach company devices and known categories of software. They do not reach a personal phone or a home computer, and they do not tell you why people went around the approved tool, which is why finding what is in use comes first.

A ban doesn’t remove the risk. It removes your view of it.

For leaders If your policy is a ban, ask what you can actually see today, and be honest about the answer.

Find it first

I would start with a conversation, not a scan. Ask each team lead what AI tools their people use, and what goes into them. Then check the places tools leave a trail: expense reports, software subscriptions, and the accounts people sign in with.

When I sit with a team and ask which AI tools their people use, it is always personal accounts, and a lot of the time the business is paying for them. The tools are almost always generic: people build a series of prompts in ChatGPT or Claude, run the answers through Copilot. They do not trust what comes back, so they keep to basic things. Now and then someone has built one or two tools to automate a task they used to do by hand. Either way, these are personal tools brought into a commercial environment that were never made for it. The first issue that comes to my mind is what happens to the intellectual property that goes into them.

Write down one line per tool. That list is the whole job at this stage.

And ask why. If people went around the approved tool, that gap is the thing to fix, before you write a single rule.

What to askWhat it tells you
The tool and the accountWhich tool, and is it a company account or a personal one?Whether you could ever see or switch it off
The people and the jobWho uses it, and for what?What the business is actually getting from it
The dataWhat goes in: customer, employee, financial or health records?How serious it is
What it can doDoes it only answer, or does it send, change or move things?Whether it needs a person approving
The ownerWho is responsible for it?Who gets the call when it breaks
The decisionAllow, move, engineer or stopWhat happens next
Figure 1. What each line on the list should hold

For leaders Ask before you hunt. A list you built with your people is worth more than a scan they work around.

The COO or CEO decides which data may go where

Every decision on the list turns on one question: which data may go into which tool. IT and security advise on the risk; the COO or CEO decides, because the trade is between speed and exposure, and a business owner makes it.

A long policy is a policy nobody reads. A short one gets followed. One page is enough if it names the approved tools, the data that never goes in, who answers questions, and the point where a tool becomes engineering.

Put one person in charge of the list: they keep it current and count again every month. New tools appear quietly, so the list is only true if someone keeps it true. That owner is the heart of AI governance: a named person who is accountable, not a committee.

I could be wrong about this in a regulated company, where a CISO may hold a veto the CEO cannot override.

For leaders Decide who decides which data may go where, and name the owner this week. A policy with no owner is a document, not a control.

Decide what to keep

Not every tool needs the same answer. Every line on the list gets one of four decisions.

  1. Allow. It is an approved tool on a company account, and no sensitive data goes in. Leave it alone.
  2. Move. The job is fine, the tool is not. Move it onto an approved tool and a company account.
  3. Engineer. Other people depend on it, it touches sensitive data, or it acts on its own. It becomes a project with an owner, a budget and engineers. The signs are in A Claude power user is not an enterprise AI strategy.
  4. Stop. Sensitive data is going somewhere nobody has reviewed, and nothing sound can replace it. This is the targeted ban, aimed at one tool. Say why, and replace the job if you can.

Most of what you find will probably be Allow or Move, and a blanket ban would have sent all of it out of sight.

I have seen some custom tools make an impact in a business. But across four or five engagements, not a single tool we have come across as a team has scaled without being completely rebuilt. I am careful never to say that their work is not good work. I think it is great for people to experiment in the right sandboxes, in the right environments, to prove out and pilot use cases. Resolon sets those sandboxes up properly, lets the people who know the work best build the tools with our assistance, and then we build them out and retool them, which is also the work we sell.

Having Claude is like having a high-performance workshop in your backyard. You can build a custom car in it, but you would not send your 16-year-old down the highway at 75 miles an hour in that car without a seat belt, or build 100 of them and sell them, because you would get sued.

Across four or five engagements, not a single tool we have come across has scaled without being completely rebuilt.

For leaders Keep the useful tools, move the risky ones, and engineer the few that the business now depends on.

What to do this quarter

Find it first, keep what helps, and govern the rest. You do not need a program for this. You need a list, an owner and a monthly habit.

  1. Ask before you hunt.

    Ask each team lead what AI tools their people use.

  2. List every tool.

    The tool, the account, the data, the owner and the decision.

  3. Name the owner.

    One person keeps the list current.

  4. Count again every month.

    Treat the list as a living record, not a one-time audit.

Frequently asked questions

What is shadow AI?

Shadow AI is any AI tool that employees use for work without the company’s approval or knowledge, for example a chatbot on a personal account. It can be a security risk, when customer, employee, financial or health records go into a tool nobody has reviewed. The risk depends on the data and on what the tool can do, which is why each tool gets its own line and its own decision.

Should we ban employees from using AI tools?

No. A blanket ban sends the use onto personal accounts where you cannot see it. Find what is in use, keep the useful tools, and set the line for when one needs engineering.

How do we find shadow AI in our company?

Ask each team lead which AI tools their people use and what goes into them, then check expense reports, software subscriptions and the accounts people sign in with.

What should an AI acceptable use policy include?

One page: the approved tools, the data that never goes in, who answers questions, when a tool becomes an engineering project, and the named owner.

Share this article

New each month. Follow by RSS

Senior-led, start to finish.

A senior Resolon lead is accountable for every engagement, from the first conversation to handover, with a team built for the work. For an active engagement or a new requirement, start a conversation.