Picture your office manager on a Tuesday afternoon. A customer sent a long, angry email about a bad job. She wants to answer it well, so she opens a free AI chat on her phone, pastes in the whole email (customer name, address, what went wrong) and asks for a polite reply. The reply is good. She sends it.
She did nothing malicious. She saved twenty minutes. But she also copied a customer's details into a tool you never picked, on a personal account, under terms nobody at your company has read.
That's "shadow AI": staff using AI tools the business hasn't approved or doesn't know about. It's the AI version of the guy who keeps customer files on a personal thumb drive. This post covers how common it is, what's actually at risk, and a one-page policy you can copy. Nobody needs to be in trouble for it to work.
How common is it?
More common than most owners guess. A few numbers, with their limits.
- The U.S. Chamber of Commerce Foundation and Ipsos surveyed 1,070 workers at small businesses (2 to 499 employees) in May 2026. Half said they already use AI at work. In 19% of cases, workers said adoption was driven mostly by employees exploring on their own. Only 11% said it was driven mostly by guidance from the organization.
- The Pax8 SMB AI Pulse Report for Q3 2026 surveyed 1,150 small and midsize business leaders in four countries. It found 35% of AI-engaged businesses had discovered shadow AI in the past six months. Pax8 sells AI software through IT providers, and the figure blends four countries, so treat it as a signal, not a precise rate.
- Vendor surveys reported in trade coverage put it higher. According to coverage in CIO.com, BlackFog research found 49% of workers use AI their employer hasn't approved. We haven't read that study ourselves, so we're not leaning on the number.
The pattern is the same across all of them: people try AI on their own because it helps, and the business finds out later. That's not a character flaw. In a business with no AI rules, trying it is the reasonable thing to do.
What's actually at risk
Most of the fear online is vague. Here's the specific version.
1. Customer and business data leaves your control. Whatever gets pasted goes to the AI company. What happens next depends on the plan and the settings. Free personal accounts and business plans often have different data terms. Our post on whether business data is safe in ChatGPT walks through what the vendors publish, as of November 2026.
2. Wrong answers reach customers. AI can sound sure and be wrong. If a staff member sends an AI-written quote, warranty answer or policy line without checking it, the mistake goes out under your name. The customer won't care that a chatbot wrote it.
3. You can't fix what you can't see. If three people use three different free tools, you can't turn off access when someone leaves, you can't tell which tool has your customer list, and you can't answer a customer who asks where their information went.
4. Known weak spots in how these tools work. The OWASP Top 10 for LLM Applications (the security community's standard list for AI systems) names sensitive information disclosure and prompt injection near the top. Prompt injection means hidden instructions in a document or web page that trick an AI into doing something the user didn't intend. The practical lesson for a small shop: don't let an AI tool act on its own, and don't feed it files you wouldn't forward to a stranger.
We didn't find data on how often shadow AI leads to a real loss at small companies. The risk is exposure and sloppiness, not a known disaster rate. That's still enough reason to write the rules down.
Why banning it doesn't work
The first instinct is "no AI at work." But the tools run in a browser tab and on a phone, where you can't see them. People who save an hour a day won't give it back because of a memo. They'll keep going and stop mentioning it. The Chamber data points the same way: more people are exploring on their own than are being guided. The gap to close is guidance.
The Intuit 2026 AI Impact Report lists privacy and security, fear of errors, and uncertainty about capability among the top barriers to adoption, not cost. Owners already feel the risk. What's missing is a clear, short answer for staff to follow.
A one-page AI use policy you can copy
Keep it to a page. If it's longer, nobody reads it. Here's a structure that works. Change the names and examples to fit your business.
1. Which tools are approved. Name them. "We use [tool] on the company business account. Ask before using anything else." One or two tools is plenty to start. If you're unsure which to pick, see ChatGPT, Claude or Gemini for a small business.
2. Use company accounts, not personal ones. Work goes through accounts the business owns, so you can remove access when someone leaves.
3. Never paste these. Make the list concrete:
- Customer names with phone numbers, addresses or account details
- Payment card numbers, bank details, Social Security numbers
- Passwords and login links
- Signed contracts, bids and anything under a nondisclosure agreement
- Employee records
- Any record your industry requires you to protect
If in doubt, remove the names or ask. "Rewrite this email, customer name removed" works fine.
4. A person checks everything that goes out. AI can draft. A human reads it before it reaches a customer, a bid or a public post. The sender is responsible for what they send.
5. Don't let AI make promises. No prices, deadlines, refunds or warranty terms go out on an AI's say-so. A person decides those.
6. Say so when it matters. If a customer asks whether they're talking to AI, answer honestly.
7. Who to ask, and what happens if something slips. Name one person. Tell staff that reporting a mistake ("I pasted something I shouldn't have") is safe, and that hiding it is the problem.
8. Review date. Put a date on the page and look at it again in six months. Tools and terms change.
We built a free AI use policy generator that turns these sections into a draft with your business name and tools filled in. It's a starting point, not legal advice. If you handle regulated data, have your lawyer read the final version. Our AI compliance guide covers the wider picture: recording laws, texting rules and vendor terms.
Make it stick
A policy in a drawer does nothing. Hold a 20-minute team meeting, walk through the page with real examples, and ask what people already use. Don't punish honest answers, or you'll never hear them again. Give people an approved tool that's at least as good as the free one on their phone. Training matters here: see how to train your team on AI. Then ask again every quarter and update the list.
Frequently asked questions
What is shadow AI?
It's AI tools that employees use for work without the business approving or even knowing about them, often on personal accounts. A common example is pasting a customer email into a free chatbot to draft a reply. The tool works fine. The problem is that nobody has checked where the data goes.
Should I ban ChatGPT at work?
A ban is hard to enforce because the tools run in any browser or phone. Surveys such as the Chamber Foundation and Ipsos study suggest staff are exploring AI on their own more than they're being guided. A short approved-tools list and a never-paste list usually works better. Our guide to ChatGPT and business data covers plan differences.
What should never go into an AI tool?
Customer details tied to a person, payment and bank information, Social Security numbers, passwords, signed contracts, employee records and anything covered by an agreement or industry rule. When staff need help with a message that contains those, remove the identifying details first. The one-page policy above lists these in plain words.
Does a small business really need a written AI policy?
If anyone on your team uses AI, yes. One page is enough. It tells staff what's allowed, protects customer information, and gives you something to point to when a mistake happens. Use the free AI use policy generator to draft one in a few minutes, then adjust it.
How Di-Hy fits
The policy is the easy part. The hard part is finding out what your team already uses, picking the right approved tools, setting up the accounts and training people to use them well. That's what our free AI audit starts with: a look at your real workflow, in plain English, and a list of what's safe to turn on. We then help you set it up, train your team, and keep it running as tools and terms change. For the bigger picture, read AI implementation for small business. We're based in Ybor City and work with businesses across Tampa Bay.
Next step: ask your team, today, which AI tools they've used for work this month. Write the answers down. That list is the first draft of your policy.
General information, not legal advice. Survey figures are from vendor and nonprofit surveys read in November 2026 and have the limits described above.


