By Don Oxman and Ronin Beaver

Here’s an uncomfortable question: how many of your employees used AI to do their jobs last week? If your answer is “none” or “I’m not sure,” that’s the problem, because the honest answer in almost every business is “more than you think.”

AI didn’t arrive in your company through a purchase order. It arrived through a web browser tab. Someone pasted a client email into ChatGPT to draft a reply. Someone dropped a spreadsheet into a free tool to summarize it. Someone ran an AI notetaker nobody vetted on a client call. None of it went through IT. None of it went through you. That’s shadow AI, and it’s now one of the fastest-growing risks in business.

The numbers are stark. In LayerX’s 2025 enterprise report, roughly 45 percent of employees were using generative AI tools, and 77 percent of those users had pasted data straight into a chatbot—with about a fifth of those pastes containing personal or payment information. Once your data lands on a third party’s servers, you no longer control where it goes, whether it trains a model, or who eventually sees it.

This isn’t just a privacy headache. IBM’s 2025 Cost of a Data Breach Report found that one in five breaches now involves shadow AI, and those breaches cost roughly $670,000 more than the rest. You end up paying a premium for a tool you never approved.

And it’s everywhere, because it looks different in every corner of the company. In sales, it’s a customer list pasted in to sharpen a pitch. In HR, it’s resumes and candidate notes. In finance, it’s internal numbers dropped in for a quick summary. In legal, it’s contract language. Same convenient shortcut, same sensitive data slipping out through the same open door.

Data leakage is only the first risk. Unchecked AI also hands your team confident, well-written answers that are sometimes flatly wrong, and an employee who trusts them can drop a hallucinated number into a proposal or a made-up citation into a contract. It creates a new attack surface, too. AI tools can be manipulated through the very content they’re asked to read, and every AI vendor you touch is a company holding your data and an account that can be breached.

The tool that makes you faster today can stop your business cold tomorrow. 

But there’s a second problem hiding underneath the first, and it’s the one owners rarely see coming. The same unchecked adoption that leaks your data also builds invisible dependencies. When a process quietly starts running through an AI tool—a workflow, a customer-facing chatbot, an automation nobody documented—that tool becomes a single point of failure. And AI tools fail. They have outages. They get breached. They change overnight, or a vendor pulls a feature. If your operation leans on one and it disappears one morning, what happens to your business?

This is where the two halves of the AI conversation meet. The risk of adopting AI carelessly and the risk of not being able to operate without it are the same risk seen from two sides.

The fix isn’t to ban AI. Bans don’t work. People route around them, and you’d be handing the advantage to competitors who use the technology well. The fix is to govern it, and that starts with three moves.

• First, find it. You can’t protect or plan around what you can’t see. Inventory where AI is actually being used across your business—the sanctioned tools and the shadow ones alike. Ask your team directly. Most people will tell you if you make it safe to answer honestly.

• Second, put guardrails on it. Write a plain-language acceptable-use policy that spells out what data can and can’t go into which tools. Move people onto business-tier or enterprise AI accounts that contractually don’t train on your data and that sit inside your security controls, instead of the free personal account. Many businesses still have no AI-specific rules, so clearing that bar puts you ahead of the field.

• Third, plan to operate without it. Treat every AI tool your business depends on the way you’d treat any critical vendor. Keep a copy of the underlying data and the process outside the tool. Know the manual fallback for anything that touches a customer. Picture your service desk running through an AI assistant that goes dark during your busiest hour: if the team has no way to pick up the work by hand, a vendor’s outage quietly becomes your outage, and your customers are the ones who feel it. Fold AI into your incident response and business continuity plan, so that “the AI is down or compromised” is a scenario you’ve already rehearsed rather than one you’re discovering live, in the middle of a bad week.

None of this requires you to become a technologist. It requires you to treat AI as what it actually is: a powerful vendor with access to your data and a growing role in how you operate. You’d never let an unknown contractor wander your building with a key to the filing cabinet. Unchecked AI is that contractor.

The businesses that come out ahead in 2026 won’t be the ones that avoided AI, and they won’t be the ones that adopted it fastest. They’ll be the ones that can answer three questions without flinching. Where is AI in our business? What are the rules? And if it broke tomorrow, could we keep running?

Don Oxman (left) is the founder of Total 360 Security, a Texas-based virtual CISO and cybersecurity firm. Ronin Beaver is a support specialist who specializes in security-focused development and networking. Learn more at total360security.com.