What Poor AI Governance Actually Costs a Small Business

Poor AI governance costs a small business in four measurable ways: breaches that cost more and take longer to contain, insurance coverage that is narrowing as carriers add artificial intelligence exclusions, regulatory exposure that is arriving faster than most policies are being written, and contracts that stall when a client asks how you control AI and nobody can answer. IBM’s 2026 Cost of a Data Breach Report found that 92 percent of organizations that suffered an AI-related security incident lacked adequate AI access controls. The tools were in use. The governance was not. 

Key takeaways

• 45 percent of employees now use AI on corporate devices, authorized or not, up from 15 percent a year earlier, according to the 2026 Verizon Data Breach Investigations Report.

• Shadow AI appeared in 43 percent of breaches in IBM’s 2026 report, more than double the prior year.

• The Insurance Services Office issued commercial general liability AI exclusions effective January 2026, and some carriers have adopted them.

• Underwriters have moved past asking whether you use AI. They now ask which models you use and what controls make the output verifiable.

• AI governance belongs in the operating budget as a recurring line item, not in a project plan that competes for capital each year.

How much AI is already running inside the average business? 

More than most leadership teams realize, and most of it was never approved. The 2026 Verizon Data Breach Investigations Report found that 45 percent of employees are now regular AI users on corporate devices, whether or not that use is authorized, a sharp rise from 15 percent the year before. It also found that 67 percent of users access AI services from non-corporate accounts on those same corporate devices. 

That distinction matters more than the raw adoption number. An AI tool reached through a personal account sits outside your admin console, outside your retention policy, and outside your ability to revoke access when that person leaves. Verizon now ranks shadow AI as the third most common non-malicious insider action in data loss prevention records, a fourfold increase, with source code the most frequently submitted data type. 

The financial consequence shows up in IBM’s numbers. In the 2026 Cost of a Data Breach Report, the share of security incidents involving shadow AI more than doubled in a year to 43 percent of breaches. The prior year’s report found that heavy shadow AI use added roughly $670,000 to the average cost of a breach and that 63 percent of breached organizations either had no AI governance policy at all or were still in the middle of writing one. 

Why does AI keep entering the business without approval? 

Because in nearly every case, someone was trying to do their job faster, and the approved path was either missing or closed. 

This is the part that gets misread as a discipline problem. It usually is not. A marketing coordinator summarizing a long report, a bookkeeper drafting a client email, or a front desk employee working through a queue of messages is not trying to evade a policy. They are trying to finish the work in front of them. When the company has not provided a sanctioned tool, or has provided one with a usage cap that runs out during a deadline week, the work does not stop. The tool changes. 

That is why AI governance is a budget conversation before it is a policy conversation. A written policy that staff cannot follow during their busiest week is not a control, it is a document. If a license cap is the reason somebody opened a personal account at four o’clock on a Friday, then the cap is the governance failure, and no amount of policy language will fix a capacity problem. 

What does effective AI governance look like for a 25 to 75 person firm? 

It looks like seven things, none of which require a dedicated compliance team. The structure below follows the four functions in the National Institute of Standards and Technology AI Risk Management Framework, which organizes AI risk work into Govern, Map, Measure, and Manage, and which is free to use. 

• Inventory what is actually running. Pull the connected applications and OAuth grants from Microsoft 365 or Google Workspace, review installed browser extensions, and list the AI features already switched on inside software you pay for. This list is almost always longer than anyone expects. 

• Classify your data before you classify your tools. Three categories is enough: information that never goes into an AI tool, information that goes only into sanctioned tools, and information that can go anywhere. Most firms have never drawn that line, which is why staff cannot follow it. 

• Put access controls on AI the way you do on everything else. IBM found that 92 percent of organizations with an AI-related incident lacked adequate AI access controls. This is the single clearest gap in the data, and it is a configuration problem rather than a cultural one. 

• Provision for peak weeks, not average weeks. Size the sanctioned option so that nobody has a practical reason to route around it. This is the line item most often left out of the budget, and it is the one that quietly determines whether the rest of the program works. 

• Write a short acceptable use policy and say it out loud. One page naming approved tools, prohibited data, and who to ask. It takes about an hour, and it is the artifact an underwriter, an auditor, or a client will ask to see. 

• Log and monitor what AI tools are asked to do. If you cannot reconstruct what a tool was given and what it returned, you cannot investigate an incident involving it, and you cannot demonstrate control to anyone who asks. 

• Name an owner and a review date. AI features are added to existing software continuously. A program reviewed once a year is a program that is wrong for eleven months. 

What should AI governance cost in the budget? 

Less than most executives assume, and the comparison that matters is not cost against zero. It is cost against the exposure it removes. 

Treat AI governance as a recurring operating line rather than a capital project. For a firm of this size it generally consists of sanctioned tool licensing sized for peak demand, the configuration work to apply access controls and logging, a modest allocation for policy and training time, and a periodic assessment to confirm that what you think is running is what is actually running. None of those are large numbers on their own. All of them are smaller than the alternative. 

What happens when AI is left out of your cybersecurity and GRC plan? 

Four things, and they compound. None of them announce themselves at the time the decision is made. 

1. Breaches cost more and take longer to contain 

IBM’s 2026 report put the global average cost of a data breach at $4.99 million, a record and a 12 percent increase over the prior year, with the United States average at $11.5 million. Incidents involving AI averaged more than $6 million, with AI-driven attacks adding roughly $1 million to the average. Model inversion attacks averaged $6.07 million and prompt injection attacks averaged $5.89 million. 

The time dimension is worse. The average breach lifecycle reached 247 days, reversing several years of gradual improvement. Breaches that ran past 200 days averaged $5.65 million against $4.32 million for those resolved faster. Ungoverned AI lengthens that clock directly, because an incident involving a tool nobody documented cannot be scoped quickly. 

2. Insurance coverage is narrowing while you wait 

This is the development most executives have not priced in yet. The Insurance Services Office issued artificial intelligence exclusions for commercial general liability policies, numbered CG 40 47, CG 40 48, and CG 35 08, effective January 2026, and some carriers have begun adopting them. Other carriers have moved the opposite direction and are offering affirmative AI coverage through endorsements, which means the market is actively splitting. 

Underwriters have changed what they ask. According to Business Insurance, they have moved past asking whether a business uses AI at all and now ask which models the company uses, how it made business decisions about those models, and what controls ensure the models produce accurate, verifiable results. A firm that cannot answer those questions is not simply paying more. It is negotiating from a position where the carrier assumes the worst. 

Marsh chief executive John Doyle warned in October that broad AI exclusions could create real confusion and uncertainty for buyers, which is a reasonable summary of where the market sits. The practical takeaway for a 40 person firm is simple. Your renewal is now partly an exam on your AI controls, and the answers are due before the quote. 

3. Regulation is arriving, and Virginia is on the list 

Virginia’s High-Risk Artificial Intelligence Developer and Deployer Act passed the General Assembly in 2025 and was vetoed that March as too burdensome. A successor, Senate Bill 365, was continued to the 2027 session. Separately, House Bill 797 directed the Joint Commission on Technology and Science to evaluate a framework for independent verification of AI models and applications, with its report due November 1, 2026. 

Nothing there binds a Virginia business today, which is exactly the point. Writing an AI acceptable use policy now costs about an hour of leadership time. Writing one under a compliance deadline, with counsel reviewing it, costs considerably more than that. 

4. Clients start asking before they sign 

Third parties were involved in 55 percent of small and medium business breaches in the 2026 Verizon report, and breaches involving third parties rose 60 percent to 48 percent of all breaches. Your clients have read the same research. For professional services firms in particular, where the client is handing over financial records, legal files, or patient information, the question of how you govern AI is moving into the diligence conversation and into contract language. 

At that point the cost of poor AI governance stops being theoretical. It is a deal that slows down, a questionnaire you cannot complete, or a renewal that goes to a competitor who can. 

Where to start 

Start with what is actually running, because every other decision depends on it. Pull the connected applications list for your Microsoft 365 or Google Workspace tenant and read it end to end. It takes an afternoon, and it converts an abstract worry into a specific list of names, permissions, and dates. 

If you would rather not do that alone, or you want the findings mapped against the rest of your security posture, that is what a risk assessment is for. 

Technology Assurance Group works with small and midsize businesses across Richmond and central Virginia to evaluate where security is weakest, build a prioritized roadmap against your requirements, goals, and budget, and implement the fixes from start to finish. Schedule a risk assessment with TAG and find out what is already running inside your business before someone else does. 

For a closer look at the tooling side of this, read our guide to AI risk management software for SMB leaders. 

Frequently asked questions 

What is AI governance? 

AI governance is the set of policies, controls, and oversight practices that determine which AI tools a business approves, what data is allowed near them, who can access them, and how their use is monitored. For a small business it is usually a one page policy, a tool inventory, access controls, and a review schedule rather than a formal program. 

How much does poor AI governance cost? 

IBM’s 2026 Cost of a Data Breach Report put the global average breach cost at $4.99 million and found AI-related incidents averaging more than $6 million. The prior year’s report attributed roughly $670,000 in additional breach cost to heavy shadow AI use. For a small business the more immediate costs are usually insurance friction, stalled client diligence, and extended downtime. 

What is shadow AI? 

Shadow AI is any AI tool used for company work without the organization’s approval or oversight. It includes free consumer chatbots, AI accounts created on work email outside of procurement, and AI features switched on by default inside software the business already uses. 

Does a small business really need an AI acceptable use policy? 

Yes, and increasingly for commercial reasons rather than only security ones. Cyber insurance underwriters now ask about AI controls at renewal, and clients in regulated industries are adding the question to vendor diligence. A one page policy is the artifact both audiences ask to see. 

Where should AI governance sit in the budget? 

In the recurring operating budget alongside other security and IT spending, not in a one time project line. The largest single component is usually licensing for sanctioned tools sized for peak demand, because under-provisioning is the most common reason employees move work to unapproved tools. 

Sources 

IBM Cost of a Data Breach Report 2026, key findings, eSecurity Planet 

IBM 2026 Data Breach Report: 92% of AI incidents had no access controls, Cybersecurity Insiders 

2026 Data Breach Investigations Report, executive summary, Verizon 

Insurers, brokers adjust as AI exclusions emerge, Business Insurance 

Broad AI exclusions in insurance policies could create uncertainty, Marsh CEO says, Insurance Journal 

Artificial Intelligence Risk Management Framework, AI RMF 1.0, National Institute of Standards and Technology 

The IBM 2026 breach report and the AI governance gap, Kiteworks