Why Shadow AI Happens and How Leaders Bring It Into the Open

Unsanctioned AI use maps your team’s unmet needs, and a crackdown only drives it deeper.

Unsanctioned AI use maps your team’s unmet needs, and a crackdown only drives it deeper.

Shadow AI is the use of AI tools, models, or AI features inside everyday software by employees without the knowledge or approval of their organization’s IT, security, or leadership teams. It usually starts with good intentions: someone finds a tool that saves them an hour a day and uses it before anyone has said yes. For most leaders, the real question is not whether shadow AI exists in their organization, but what it is telling them and what to do about it.

What Shadow AI Is and Why It Matters

Shadow AI is a close cousin of shadow IT, the long-standing pattern of employees adopting software that IT never vetted. The difference is speed and reach. A shadow IT tool usually needed a credit card and an account. A shadow AI tool often needs nothing more than a browser tab, or it arrives automatically as a new AI feature switched on inside a SaaS product the company already pays for. The scale is not small. The 2024 Work Trend Index from Microsoft and LinkedIn (May 2024\) found that 75% of knowledge workers use AI at work and that 78% of AI users bring their own AI tools to work. A Software AG study of 6,000 knowledge workers in the US, UK, and Germany (reported October 2024\) found that roughly half use unsanctioned AI tools, and 46% said they would keep using them even if their employer banned them. The risk is real too. IBM’s 2025 Cost of a Data Breach report (July 2025\) found that one in five organizations reported a breach due to shadow AI, and organizations with high levels of shadow AI saw an average of $670,000 in higher breach costs than those with low or no shadow AI. The same report found that only 37% of organizations have policies to manage AI or detect shadow AI. Put those findings side by side and a pattern emerges. Most employees are already using AI. Many are doing it outside approved channels. And most organizations do not have the visibility to know where.

Why Shadow AI Happens

Leaders often read shadow AI as a discipline problem. In practice it is almost always a supply problem. People reach for unapproved tools when the approved path is slower, weaker, or nonexistent. There are four common causes.

1\. The approved tools do not fit the work

An organization may have rolled out one enterprise assistant, but a product manager needs help synthesizing customer interviews, a developer wants a coding assistant inside their editor, and a marketer needs image generation. When the sanctioned tool does not match the job, people find one that does.

2\. The approval process is slower than the need

If getting a new AI tool reviewed takes three months and the deadline is Friday, the deadline wins. Long procurement and security reviews were designed for multi-year software contracts, not for tools that change every few weeks.

3\. Nobody has said what is allowed

Many employees are not hiding anything. They simply have never been told where the lines are. Without clear guidance, people make their own judgment calls about what data is safe to paste into a chat window, and those calls vary widely.

4\. AI is arriving through tools you already own

A growing share of shadow AI is not a new app at all. It is an AI feature that a vendor switched on inside a CRM, a document editor, a meeting recorder, or a design tool. No one downloaded anything, and no one asked for approval, because from the employee’s point of view nothing new was installed.

Shadow AI as a Signal, Not Just a Risk

Here is the reframe that changes how leaders respond: every instance of shadow AI is data about where your workflows are broken and where your sanctioned options fall short. If a dozen people on the sales team are quietly using the same note-taking assistant, that is a strong signal about a real need. If engineers are running code through a personal assistant, that tells you something about the gap between your approved tooling and the way modern development actually works. If an AI transformation leader treats those findings as violations only, they throw away the most honest map of demand they will ever get. This is also where human AI collaboration actually begins. The people using shadow AI are, in many cases, the early experimenters your AI strategy depends on. They have already figured out where AI helps and where it does not. Pushing that behavior underground loses their knowledge and their goodwill at the same time. None of this means ignoring the risk. Sensitive customer data, source code, and confidential plans should not be flowing into tools nobody has vetted. The point is that the fastest way to reduce that risk is to make the safe path the easy path, and you cannot design the easy path without knowing what people are actually trying to do.

Shadow AI vs. Sanctioned AI: Key Differences

	Shadow AI	Sanctioned AI
Who chose it	Individual employee or team	IT, security, or leadership
Visibility	Low or none	Logged and monitored
Data handling	Unknown terms, possible training on inputs	Reviewed contracts and data controls
Speed to adopt	Immediate	Often weeks or months
Fit to the task	Usually high, chosen for a specific job	Varies, often general purpose
Support and training	None	Available

The table makes the tension clear. Shadow AI usually wins on fit and speed. Sanctioned AI wins on safety and visibility. A good governance approach tries to close both gaps instead of picking one.

Common Pitfalls When Responding to Shadow AI

Banning everything

Blanket bans are the most common first reaction and the least effective. The Software AG finding that 46% of knowledge workers would keep using their tools after a ban suggests what happens next: usage continues; it just becomes less visible. A ban trades a visible problem for an invisible one.

Treating discovery as an investigation

If the first step is a monitoring tool and a list of names, people will stop telling you anything. Leaders who want an honest picture of AI use have to make it safe to share. Fear produces silence, not compliance.

Writing a policy and stopping there

A written policy matters, and it is a necessary part of the answer. But a policy on its own does not tell people which tools they can use tomorrow morning. Without an approved list and a fast path to add to it, the policy describes rules nobody can follow.

Ignoring AI features inside existing software

Organizations often audit new apps while missing the AI features already turned on inside their existing stack. A complete picture includes vendor features, browser extensions, and personal accounts, not just standalone tools.

Leaving it entirely to IT

Shadow AI is not purely a security problem. It touches workflow design, team habits, training, and strategy. When it sits only with IT, the response tends to be all controls and no enablement. The best responses involve IT, security, legal, and the business leaders whose teams are doing the work.

How Leaders Bring Shadow AI Into the Open: 5 Steps

The following five steps work for a single team or an entire organization. They are written for a Director or VP who wants to reduce risk without killing the experimentation they say they want.

Step 1: Announce amnesty before you ask questions

Start by making it explicitly safe to talk about what people are using. Say plainly that the goal is to learn, not to punish, and that nobody will be disciplined for tools they disclose. This one message determines whether everything that follows produces honest answers.

Step 2: Run a no-blame AI inventory session

Bring teams together in a facilitated working session and ask three questions: What AI tools are you using, including features inside software we already own? What job does each one do for you? What data goes into it? Capture the answers on a shared board grouped by job to be done, not by person. In Voltage Control’s facilitated AI working sessions, this format consistently surfaces far more tools than a survey does, because people build on each other’s answers and realize their workaround is not unusual. A survey can supplement the session, but the live conversation is where the “why” comes out, and the “why” is what tells you how to respond.

Step 3: Sort what you find into three buckets

Once you have the inventory, sort each tool or use case into one of three groups:

• Approve: Low risk, high value. Move it onto the sanctioned list quickly, with any needed settings or contract terms.
• Replace: The need is real, but the tool is not acceptable. Find or provide a sanctioned alternative that does the same job, and say when it will be ready.
• Stop: The risk outweighs the value, usually because of the data involved. Explain why clearly, and point people to the closest safe option.

The key is that “stop” should be the smallest bucket. If most of what you find ends up there, people will go right back underground.

Step 4: Create a fast, lightweight vetting path

A useful model is to govern AI tools the way you onboard a new hire. You would not give a new employee access to every system on day one, and you would not make them wait six months to start either. You check references, set a clear scope, give them access to what the role needs, and expand trust as they prove themselves. Applied to AI tools, that means a short intake form, a defined review owner, a target turnaround measured in days rather than months, and a starter scope (for example, approved for public or internal data, not for customer data) that can be widened after a trial period. When the vetting path is fast, there is much less reason to go around it.

Step 5: Make it ongoing, not a one-time cleanup

AI tools change too quickly for a single audit to stay accurate. Build a regular rhythm: a quarterly inventory refresh, a visible channel where people can request new tools, and a short review of what was added or retired. Treat AI tool requests the way an AI product manager treats a roadmap, as a running backlog of needs to prioritize, not a stack of exceptions to deny.

Where Policy Fits

Bringing shadow AI into the open and writing an AI acceptable use policy are related but different jobs. The inventory and vetting process tells you what people need and what risks are actually present. The policy turns those findings into clear, durable rules. Organizations that write the policy first often end up rewriting it once they learn what is actually in use. Doing discovery first gives the policy something real to govern. It also helps to connect this work to the broader AI strategy. Shadow AI is often the first honest data point on where AI can create value in your organization. Teams that are already experimenting can become pilot groups, internal champions, or the source of use cases for a broader rollout. For more on how this kind of change lands with people, see Voltage Control’s pieces on the new friction teams face with AI, what human AI collaboration looks like in practice, and AI-driven change management.

Frequently Asked Questions About Shadow AI

Is shadow AI always a bad thing?

No. Shadow AI carries real data and security risk, but it also reveals unmet needs and identifies the employees who are already experimenting productively. The goal is to bring that activity into the open so it can be supported safely, not to eliminate experimentation.

How do I find out what shadow AI is in use?

Combine a no-blame facilitated inventory session with a short survey and a review of AI features already enabled in your existing software. Technical monitoring can help, but it works best after you have made it safe for people to disclose what they use.

Who should own shadow AI in an organization?

Ownership works best as a shared responsibility. IT and security typically own the vetting process and controls, legal owns data and contract terms, and business leaders own the workflows and the case for which tools their teams need. Some organizations assign an AI transformation lead to coordinate across these groups.

Should we hire an AI transformation consultant to deal with shadow AI?

It depends on internal capacity. Many organizations can run an inventory and set up a vetting path themselves. An outside AI transformation consultant or facilitator is most useful when teams do not trust the process enough to be honest, when the discovery needs to span many departments, or when leadership wants to connect the findings to a broader AI strategy.

Bring Shadow AI Into the Open With Your Team

Shadow AI is not going away, and trying to stamp it out usually just pushes it out of sight. The leaders who handle it well treat it as an honest signal of what their teams need, make it safe to talk about, and build a fast path to the tools people are already asking for. Voltage Control helps leadership teams run the facilitated working sessions that surface real AI use, align IT, security, and the business on what to approve, and turn the findings into a governance approach people will actually follow. Book a free intro call with our facilitation team to plan your first shadow AI working session.