Your employees are already using AI tools you haven’t approved, don’t know about, and can’t monitor. It’s probably happening right now in a browser tab next to their email inbox. This is shadow AI — the unsanctioned use of ChatGPT, Gemini, Claude, or dozens of other AI tools to get work done faster — and for most small and mid-sized businesses, it’s already bigger than anyone realizes.

Shadow AI isn’t a far-off threat. It’s a current-state reality that’s quietly reshaping where your sensitive data lives, who can see it, and how exposed your business really is.

How Widespread Is Shadow AI, Really?

The numbers are striking no matter which research firm you look at. 8 in 10 employees use unauthorized AI tools and 68% of security leaders admit to using unauthorized AI in daily work, according to UpGuard’s State of Shadow AI report. That second stat is the one worth sitting with — the people responsible for stopping shadow AI are often doing it themselves.

The trend is also accelerating fast. Verizon’s 2026 Data Breach Investigations Report found that employee use of unapproved AI tools tripled in a single year, from 15% to 45% of employees. That kind of growth curve means a policy written even a year ago is probably already out of date.

It’s not limited to casual use, either. Fifty-one percent of employees have connected or integrated AI tools with other work systems or apps without IT approval or oversight, which means shadow AI isn’t just a chatbot in a browser tab — it’s quietly wiring itself into your business systems.

What Employees Are Actually Putting Into These Tools

The adoption numbers matter less than what’s being typed into the prompt box. Employees routinely paste in exactly the kind of information you’d expect a business to protect.

• Twenty-seven percent of employees using unapproved AI have entered employee data into these tools.
• Twenty-three percent of people have shared internal financial statements or sales data with unsanctioned AI.
• Twenty-eight percent of data-loss events involving shadow AI contained source code.

None of this usually comes from bad intent. An office manager drafting a layoff memo, a salesperson summarizing a client contract, or a developer debugging a snippet of proprietary code is just trying to move faster. But once that information is pasted into a third-party AI tool, your business has effectively lost control of where it goes, how long it’s retained, and who else might be able to access it.

Why It Costs More When Something Goes Wrong

This is where shadow AI stops being a productivity question and becomes a budget question. IBM’s Cost of a Data Breach Report 2025, based on interviews tied to 600 breached organizations, put a hard number on the risk for the first time.

IBM’s Cost of a Data Breach Report 2025 found that 20% of breached organizations were compromised through shadow AI, the unsanctioned generative AI tools employees adopt without security sign-off, and those incidents added roughly $670,000 to the average breach. That’s not a rounding error on a budget line — for a lot of SMBs, that’s close to the entire annual IT spend.

The breach data gets worse before it gets better. Breaches involving shadow AI also have longer lifecycles (247 days vs. the global average), higher rates of customer PII compromise (65% vs. 53% overall), and increased intellectual property theft (40%). Longer lifecycle means longer exposure — more time for attackers to move around before anyone notices.

Perhaps the most telling number in the whole report: shadow AI added an average of $670,000 to breach costs at organizations with high unauthorized AI use, according to IBM’s 2025 research. Nearly all of those incidents shared one root cause. Ninety-seven percent of AI-related breaches lacked proper access controls. In other words, this isn’t an unsolvable problem — it’s an ungoverned one.

Why “Just Ban It” Doesn’t Work

The instinct for a lot of business owners is to block AI tools outright at the network level. It’s understandable, but it rarely holds up in practice. Employees who find a tool genuinely useful will often find a way around the restriction — using a personal phone, a home computer, or a personal account on a work device. That shift actually makes things worse, because now the activity happens completely outside your visibility, with none of the logging or oversight you’d have on a managed device.

A flat ban also ignores why shadow AI took hold in the first place: employees are trying to solve real work problems and the company hasn’t given them an approved way to do it. The more sustainable path is to meet that need with a sanctioned alternative, paired with clear rules about what can and can’t be shared.

Building a Practical Shadow AI Policy

You don’t need a 20-page document to get this under control. A workable policy covers a handful of core points:

• Name the approved tools. If your business already pays for Microsoft 365 Copilot or another enterprise AI tool with business-grade data protections, make that the default option employees are pointed to.
• Define what can’t be shared. Spell out categories — customer records, financial data, source code, anything covered by compliance obligations — that should never go into a public AI tool, sanctioned or not.
• Explain the why, not just the rule. Employees who understand that pasted data can be retained or used to train external models are far more likely to follow the policy than those who just see a restriction.
• Review access regularly. New AI features get added to existing software constantly. A tool that was harmless six months ago may now have a built-in AI assistant with access to your files.
• Monitor what you can. Network and endpoint monitoring tools can flag traffic to known AI platforms, giving you visibility even without a blanket ban.

This kind of governance isn’t unique to AI — it’s the same discipline that goes into automated threat detection and response in cybersecurity defense more broadly: visibility first, then controls, then response.

Where to Start This Week

If your business doesn’t have an AI usage policy yet, you’re in the majority, not the minority. In IBM’s Cost of a Data Breach Report 2025, breaches at organizations with high shadow-AI use cost $670,000 more on average — and one in five surveyed organizations reports a breach caused by shadow AI. That gap between “most companies have no policy” and “the cost is already measurable” is exactly why this is worth addressing now rather than after an incident forces the issue.

Start small: ask your team what AI tools they’re already using and why. That single conversation usually surfaces more shadow AI activity than any monitoring tool will catch on day one, and it gives you a realistic starting point for a policy people will actually follow.

If you’re not sure how exposed your business is or where to start writing a policy that fits how your team actually works, a managed IT partner can help assess your current AI usage and put reasonable guardrails in place — without shutting down the tools that are genuinely helping your team get work done.