Generative AI has become a regular part of how employees work, often well ahead of any formal policy. Research finds that the average organization now experiences 223 incidents per month of employees sending sensitive data to AI tools, double the rate from a year ago. Shadow AI breaches carry an average $670,000 premium over standard data breaches. This post explains what ungoverned AI use actually looks like inside a small business, why the risk is real, and what a practical governance approach looks like without overcomplicating it.
Ask most business owners whether their team is using AI tools at work and the answer is usually yes. Ask whether there’s a formal policy governing how those tools are used and the answer is usually no, or not really.
That gap is where the risk lives.
Generative AI tools like ChatGPT, Gemini, Copilot, and dozens of others have spread through workplaces at a pace that governance simply hasn’t matched. Employees use them to draft emails, summarize documents, research topics, clean up data, and solve problems faster. Most of them aren’t doing anything they’d consider unusual. They’re just trying to get their work done.
The problem isn’t the intent. It’s what flows into those tools along the way.
According to research tracking AI usage across organizations, the average company now experiences 223 incidents per month of employees sending sensitive data to AI applications, a figure that has doubled in the past year. IBM’s 2025 Cost of a Data Breach Report found that shadow AI breaches cost organizations $670,000 more than standard breaches, and shadow AI now accounts for 20% of all data breaches tracked.
For a small business, a breach at that scale isn’t an inconvenience. It’s a genuine threat to operations, client relationships, and reputation.
What Ungoverned AI Use Actually Looks Like
The Shadow AI Problem Isn’t New, But It’s Getting Bigger
The term “shadow AI” refers to AI tools that employees use without IT approval or oversight. It’s the same dynamic as shadow IT, where someone installs an app or signs up for a service outside of any formal process, except that AI tools can process, retain, and in some cases train on the data you put into them.
Research from Menlo found that 68% of employees who use AI tools at work do so through personal accounts on free platforms. Of those, 57% use sensitive data in those sessions. The personal account is the key detail: when an employee logs into ChatGPT with their personal Gmail account and pastes in a client proposal or an internal financial summary, that data moves into a system the business has no control over, no visibility into, and no way to retrieve.
A 2025 Proofpoint analysis found that 77% of employees have been observed sharing sensitive or proprietary information with tools like ChatGPT. That’s not a fringe behavior. It’s how most people who use AI tools actually use them.
It’s Not Just Standalone Chatbots
One of the most important and least understood aspects of shadow AI in 2026 is that it isn’t limited to employees signing up for external tools.
AI features are now built into software your business already pays for. Grammarly can analyze and suggest edits on documents. Zoom can transcribe meetings and summarize them. Salesforce has AI capabilities embedded in its CRM. Microsoft Copilot is rolling out across Microsoft 365. Adobe has generative AI features in its creative tools. Many of these features are enabled by default or can be switched on by individual users without any IT review.
Research from Acuvity found that 18% of organizations are specifically concerned about AI features embedded in approved SaaS applications, capabilities that are often analyzing and processing company data without anyone having made a deliberate decision to allow that.
If you haven’t reviewed which AI features are active inside your existing software stack, you likely have more AI processing your business data than you realize.
The Risk Isn’t Just Theoretical
When employees put sensitive data into an AI tool without governance controls in place, a few specific things can happen.
Data can be retained by the tool and used for model training. Many AI services make this opt-out rather than opt-in, which means it’s happening unless someone has specifically gone in and changed the settings. Business information, client data, and internal documents that get pasted into a model can become part of that model’s training data and potentially influence responses to other users.
Data can be exposed in a breach of the AI provider. The AI tools your employees use are subject to security incidents just like any other cloud service. If those tools hold data from your business, your business is exposed in the event of an incident at the vendor.
Compliance requirements can be violated silently. If your business handles client data under any kind of regulatory framework, HIPAA, state privacy laws, financial services regulations, or contractual data handling obligations, your team using unapproved AI tools to process that data may already be creating compliance exposure that you can’t see.
Our cybersecurity consulting services include reviewing AI usage exposure as part of risk assessments, because it’s one of the fastest-growing gap areas we encounter in small business environments.
Why Banning AI Isn’t the Answer
It’s tempting to respond to this by issuing a policy that prohibits AI tools. The problem is that this approach consistently fails.
Employees who are used to working with AI and find it genuinely useful don’t stop using it because of a policy. They find ways to use it that are harder to detect. The behavior goes underground rather than going away, and in doing so, becomes even harder to govern.
Research from Gartner found that 69% of organizations suspect or have confirmed evidence that employees are using prohibited public AI tools. Microsoft’s research found that 71% of UK employees admitted to using unapproved AI tools at work, with more than half doing so at least weekly. The prohibition approach isn’t working at large organizations with dedicated security teams. It’s unlikely to work at a small business without the same enforcement infrastructure.
The more effective approach is to meet employees where they are. Acknowledge that AI is part of how work gets done. Provide approved options that meet their actual needs. Make the approved path easy enough that the unapproved path isn’t worth the trouble. And back that up with clear guidance on what data can and can’t go into any AI tool, approved or otherwise.
What a Practical AI Governance Framework Looks Like
Step 1: Find Out What’s Actually Being Used
You can’t govern what you can’t see. Start by getting an honest picture of which AI tools your team is currently using, both the ones you know about and the ones you don’t.
Identity logs and network activity can surface AI-related domains your team is connecting to. A direct, nonjudgmental conversation with your team often surfaces more than any technical audit. Ask what tools they’re using, what they find useful, and what problems they’re trying to solve. The answers will tell you where to focus.
This is the same discovery process we covered in our earlier post on shadow IT and unsanctioned cloud apps. AI tools are the same problem in a new category.
Step 2: Classify What Data Can Go Where
The most practical governance control for a small business isn’t a long list of approved and prohibited tools. It’s a clear data classification framework that employees can actually apply in real time.
Keep it simple. Four categories are enough:
Public information is already available externally and poses no risk if it enters an AI tool. Using AI to summarize a news article or draft marketing copy about a public product is low risk.
Internal information is business data that shouldn’t leave the organization but isn’t highly sensitive. General business processes, non-confidential internal documents, and operational information fall here. Most AI tools should not receive this without review.
Confidential information includes client data, financial details, pricing, legal documents, proprietary processes, and strategic plans. This category should not enter any AI tool that isn’t specifically approved and governed for that purpose.
Regulated information is anything covered by legal or contractual data handling requirements. HIPAA-covered patient data, personally identifiable information under privacy laws, and financial records subject to regulatory frameworks all belong here. Strict controls apply.
When your team understands these categories and knows which bucket their data falls into, the judgment call becomes much clearer.
Step 3: Decide Which Tools Are Approved and Communicate That Clearly
Based on what you learn in discovery and what your data classification framework reveals, you can make deliberate decisions about which AI tools are appropriate for which purposes.
An approved tool for public and low-sensitivity work might be a free tier of a common AI tool, used through a work account rather than a personal one. An approved tool for anything touching confidential or regulated data needs to be a platform with appropriate data processing agreements, privacy settings reviewed, and training opt-outs in place.
The key is that these decisions are written down, communicated to your team, and actually reflect how they work rather than being an aspirational policy that nobody reads.
Step 4: Train Your Team on Why This Matters
Policy without understanding doesn’t hold up under deadline pressure. When employees genuinely understand what can happen when sensitive data enters an unmanaged AI tool, they make different decisions. When they don’t understand the risk, convenience always wins.
Our cybersecurity awareness training covers AI-specific data handling as part of broader security education. It’s one of the areas that generates the most engagement from employees, because AI is something they’re actively thinking about and using, and the risks aren’t always obvious from the outside.
Step 5: Review and Update Regularly
AI capabilities are changing faster than almost any other area of business technology. The tool that was low risk six months ago may have added new features that change its profile. A platform your team hasn’t been using may suddenly become relevant because a competitor started using it. The embedded AI features in your existing software will continue to expand.
Build a quarterly review of your AI governance framework into your operations. It doesn’t have to be a major project each time. A check-in on what’s changed, what new tools have appeared, and whether your current policy still reflects your actual risk posture is enough to stay ahead of the drift.
AI Governance Is Part of Your Security Posture
Small businesses often think of AI governance as something for larger organizations with dedicated compliance teams. The data suggests otherwise. Shadow AI incidents are happening at businesses of every size, the data that flows into those tools includes sensitive client and business information regardless of company size, and the financial consequences of a breach don’t scale down just because the business is smaller.
Getting deliberate about AI governance doesn’t require a compliance department. It requires a clear picture of what your team is using, a simple framework for what data belongs where, approved options that meet real needs, and a team that understands why the boundaries exist.
That’s a manageable project for any business that decides to make it a priority.
If you’d like help building an AI governance framework that fits how your business actually operates, reach out to the Z-JAK team. We work with small and mid-sized businesses across Louisville to put practical policies in place before the gap between adoption and governance turns into an incident.
Frequently Asked Questions
What is AI governance and why does a small business need it?
AI governance is a set of policies, practices, and controls that determine which AI tools your team can use, what data those tools can process, and how AI-assisted work gets handled within your organization. Small businesses need it because employees are already using AI tools, often with sensitive business data, and without governance there’s no visibility into what’s being shared or where it’s going. The average business now experiences hundreds of sensitive-data-to-AI incidents every month, and the financial consequences of a breach involving unmanaged AI are meaningfully higher than standard incidents.
How do I find out which AI tools my employees are already using?
Start with what your systems can already show you: identity logs, network traffic, and the admin panels of your existing SaaS platforms often surface AI-related activity. Then ask your team directly in a nonjudgmental way. Frame the conversation as an effort to support how they work, not investigate what they’re doing. Most employees using AI tools aren’t trying to create problems. They’re trying to work efficiently, and they’ll tell you what they’re using if the conversation feels safe.
What types of data should never go into a public AI tool?
Client data of any kind, financial records, employee information, legal documents, proprietary processes, pricing details, and strategic plans should not enter public AI tools that haven’t been reviewed and approved for handling sensitive data. Anything covered by a regulatory requirement, including HIPAA, state privacy laws, or financial services regulations, warrants particular caution. A simple data classification framework with four categories, public, internal, confidential, and regulated, gives your team a practical guide for making this judgment in real time.
Can we just ban AI tools to avoid the risk?
Research consistently shows that prohibition alone doesn’t eliminate shadow AI use. It drives it underground, making it harder to detect and govern. Organizations that prohibit AI without providing approved alternatives tend to end up with higher rates of unsanctioned use, not lower. The more effective approach is to acknowledge that AI is part of how work gets done, provide approved options with appropriate controls, and educate your team on what data boundaries exist and why.
What’s the difference between an AI policy and AI governance?
An AI policy is a document that states the rules. AI governance is the broader system that makes those rules real: the discovery process that tells you what’s actually being used, the data classification framework that gives employees practical guidance, the approved tool list with controls in place, the training that builds understanding, and the regular review cycle that keeps everything current. Policy without the surrounding governance structure tends to be ignored. Governance without a clear policy lacks the structure to communicate expectations.
Ready to Build an AI Governance Framework That Works for Your Business?
AI governance doesn’t have to be complicated. It has to be deliberate. The gap between how your team is using AI today and what your current policy says is where the risk lives, and closing that gap is a manageable project with the right guidance. If you’d like to talk through where your business stands and what a practical next step looks like, get in touch with the Z-JAK team today. We’re here to help.