Serving regulated mid-market businesses nationwide (888) 901-9686 · [email protected]
← Back to Insights

Shadow AI in a small office: what happens when staff paste client data into public chatbots

Employees at small firms are quietly feeding client data into public AI chatbots to save time. Most don't know those services keep the data—and most owners don't know it's happening.

ES
Elevate Solutions
June 26, 2026 · 4 min read

Someone on your team is using a publicly available AI chatbot to work faster. Maybe they're drafting a client letter, summarizing a meeting, or cleaning up a contract. This is happening in offices with five employees and offices with five hundred. The difference is that larger firms have people writing policies to stop it. You probably don't.

When employees use publicly available AI chatbots for work tasks, the client and business data they enter may be stored, reviewed, or used to train the AI provider's models—without the employee or the firm knowing. For small offices in regulated industries, a single paste of a client file into a public chatbot can create a breach, a privilege waiver, or a regulatory violation. A written acceptable-use policy and a vetted, enterprise-licensed AI tool are the minimum controls a small firm needs today.

What is shadow AI, and why does it matter in a small office?

Shadow AI is the use of AI tools that the business hasn't reviewed, approved, or secured. Unlike the shadow IT of a few years ago—a personal Dropbox account, an unapproved app—shadow AI carries a specific risk that most small-office employees never consider: the data they type into the tool may not stay private.

What actually happens to the data employees enter

Most publicly available AI chatbot services operate under terms of service that allow the provider to store, review, and use submitted content to improve their models. That policy often changes when an organization pays for an enterprise plan with a data processing agreement—and it rarely applies at all on the no-cost, unauthenticated tier most individual employees reach for first.

When a staff member pastes a client's name, a medical record, a financial summary, or case details into one of these tools, that information leaves your office network and enters a third-party server. You have no data processing agreement with that provider. You have no audit trail. You have no reliable mechanism for deletion. And you almost certainly have no idea it happened.

For firms in healthcare, law, financial services, or any field where client confidentiality is a legal or ethical obligation, this is not a theoretical problem. It is a live compliance exposure.

Why small offices carry the most risk

Large firms issue policies, run formal training, and deploy technical controls that block unapproved tools. At a six-person office, the policy is often unwritten, training is informal, and no one is monitoring the network for data leaving through a browser tab.

Employees at small firms also tend to move fast and wear multiple hats. When a paralegal is drafting a motion and a chatbot will produce a workable draft in thirty seconds, the judgment that stops a larger-firm employee—"this might violate our policy"—is less likely to kick in when there is no policy to violate.

The risk compounds because small-firm clients tend to be individuals and small businesses themselves. A data exposure at a boutique law firm or a solo financial planner is not an abstract statistic. It is a named person whose private information is now in a vendor's training dataset.

What you can do this week — no IT department required

Write a one-page acceptable-use policy

It does not need to be long. It needs to say: employees may not enter client names, client data, confidential business information, or anything covered by a nondisclosure agreement into any AI tool not approved by the firm. List the approved tools. List the prohibited ones. Have everyone sign it. Date it.

Have the conversation before the next incident

Call a ten-minute all-hands and explain the risk plainly. Most employees are not trying to cause a breach—they are trying to be efficient. Tell them directly: the chatbot keeps what you type; if that includes client information, we may have violated a law, a contract, or a professional ethics rule. That conversation takes ten minutes and costs nothing.

Use the tools you already pay for

If your firm runs Microsoft 365 Business Premium, you have access to controls that most small offices leave untouched. The platform includes data loss prevention policies that can flag or block sensitive information from leaving your environment through a browser. Sensitivity labels let you mark documents so staff know which files carry higher handling requirements. Conditional access policies can restrict which devices and applications reach your data at all.

None of this requires enterprise licensing. These features are part of what you already pay for. A dedicated team that knows your environment can configure basic data loss prevention and sensitivity label policies in an afternoon.

When staff need AI, point them somewhere safe

There are AI tools built for regulated environments—with enterprise terms of service, data processing agreements, and controls that keep your content out of model training. The goal is not to ban productivity. It is to redirect it through a channel that doesn't expose your clients.

The bottom line

Shadow AI is not a future risk. It is happening in small offices right now, driven not by bad intent but by convenience. A written policy, a direct conversation with your team, and proper use of tools you already pay for will address most of the exposure. The firms that get hurt are the ones that assume it isn't happening to them.

Elevate Solutions' security and IT advisory team delivers managed cybersecurity (MDR/MXDR), managed IT, and compliance guidance (HIPAA, SOC 2, PCI DSS) for regulated mid-market firms across Los Angeles.

Reviewed by David Faramarzi · Founder, Elevate Solutions
Share:
Next story The Small-Office Security Scorecard: Rate Your Own Setup in 5 Minutes June 26, 2026 · 4 min read