Are Your Employees Sharing Company Data With ChatGPT?

Your employees are probably already using ChatGPT at work.

They may use it to rewrite an email, summarize a document, analyze a spreadsheet, translate something or get help with a difficult task. Most of the time, that is useful.

But there is a question many companies are only starting to ask:

What are employees actually sending to these AI tools?

A salesperson could upload a spreadsheet containing customer names and prices. Someone in finance could paste payment information into a prompt. A developer could share source code or configuration data while asking for help with a bug.

None of this requires malicious intent. In many cases, the employee is simply trying to save time and get their work done faster.

That is exactly what makes the risk difficult to see.

The Problem Is Not Employees Using ChatGPT

There is an important distinction here.

Using ChatGPT, Claude, Gemini or another AI assistant at work is not automatically a security problem. These tools can genuinely help employees work faster.

The problem starts when company information is copied, pasted or uploaded to an AI service without anyone considering whether that information should leave the company's systems.

Think about a simple example.

An employee receives a long document from a customer and needs to understand it quickly. Instead of spending 30 minutes reading it, they upload the document to ChatGPT and ask: "Can you summarize this for me?"

Thirty seconds later, they have a useful summary.

But the document might contain customer names, contact details, pricing, contract terms or other confidential information.

The employee probably did not think of this as a data transfer. They were just trying to finish a task.

From the company's perspective, however, potentially sensitive information was submitted to an external AI service.

What Kind of Company Data Could Be Shared With AI?

Almost any information employees work with can eventually find its way into an AI prompt.

For a sales team, it could be a customer list, a quotation or a pricing spreadsheet. For finance, it could be billing or payment information. HR teams might work with employee records. Developers might paste source code, configuration files or technical information while troubleshooting a problem.

Other examples include contracts, internal reports, product specifications, business plans, research, intellectual property and customer communications.

Credentials can be particularly problematic. Someone troubleshooting an integration might copy a configuration block without noticing that it also contains an API key, password or another secret.

The common factor is simple: the employee is usually sharing the information because it helps the AI produce a better answer.

This Problem Has a Name: Shadow AI

You may already be familiar with the term "Shadow IT."

Shadow AI is a similar idea. It generally refers to employees using AI applications for work outside the tools, policies or visibility established by their organization.

For example, a company may provide an approved AI assistant for employees, but someone may still prefer to use a personal ChatGPT account because they are already familiar with it.

The IT team might know that employees are visiting ChatGPT, but they may have no idea what information is being submitted.

This is becoming harder to ignore as AI adoption grows.

According to Netskope's 2026 Cloud and Threat Report, the number of people using SaaS generative AI applications in the average organization tripled over the previous year. The number of prompts being sent to these applications increased sixfold.

Netskope also reported that 47% of generative AI users were using personal AI apps.

That matters because an organization may have rules for approved corporate AI tools while still having employees using services outside those controls.

How Often Is Sensitive Data Actually Sent to AI Tools?

This is not only a theoretical risk.

Netskope reported that incidents involving sensitive data being sent to generative AI applications doubled over the previous year. The average organization in its research experienced 223 generative AI data policy violations per month.

Source code accounted for 42% of the observed violations, regulated data for 32%, and intellectual property for 16%.

The report also found that half of the organizations observed did not yet have enforceable data protection policies for generative AI applications.

This highlights an important problem.

Having an AI policy is not the same as knowing whether that policy is being followed.

Can a Company See Whether Employees Are Using ChatGPT?

Yes. Detecting the use of ChatGPT and other AI services is technically possible and relatively straightforward compared with detecting what information is being shared.

Endpoint monitoring, browser monitoring and network security tools can provide visibility into the applications and websites being used on company computers.

A company could therefore answer questions such as: Which employees are using AI tools? Which services are popular? How frequently are they being used? When are they being used?

This is useful information, but it only solves the first part of the problem.

Knowing that an employee visited ChatGPT does not tell you whether they asked it to rewrite an email or uploaded a spreadsheet containing 5,000 customer records.

Those are very different situations.

Can a Company Detect Data Being Sent to ChatGPT?

This is where the problem becomes more interesting.

Technically, companies can go beyond simply detecting that an AI website was opened. Data Loss Prevention, commonly called DLP, can be used to identify certain types of sensitive information and apply rules when that information is being transferred.

Microsoft, for example, documents Endpoint DLP capabilities for managed Windows devices that can restrict sensitive information from being pasted or uploaded to AI applications.

One of Microsoft's examples involves preventing or warning a user when credit card information is pasted into services such as ChatGPT.

Depending on the technology being used, companies can look at signals such as file uploads, copy and paste activity, browser activity and known patterns of sensitive information.

This does not mean a company needs to read every ChatGPT conversation.

A better approach is to identify specific situations that may represent risk and bring those situations to the attention of the appropriate person.

Where Anomaly Detection Becomes Interesting

Traditional employee monitoring tells you what happened.

An employee used Chrome for three hours. They visited ChatGPT. They opened a particular application. They worked between certain hours.

Anomaly detection asks a different question:

Is something happening that is unusual enough to investigate?

Consider an employee who normally visits ChatGPT a few times per week. One afternoon, their AI usage suddenly increases significantly while they are also working with customer files.

That does not mean the employee stole data.

It may be completely legitimate.

But it could be a useful signal.

Another example could be an employee who has never used an AI tool before and suddenly begins spending significant time on several generative AI services. Again, this is not proof of wrongdoing. It is simply unusual behavior that a company may want to understand.

This distinction matters.

An anomaly is a reason to look closer, not a verdict.

Why Not Just Block ChatGPT?

Some companies may decide that the easiest solution is to block ChatGPT completely.

For certain environments, that may be appropriate.

For many companies, however, it creates another problem. Employees use AI because it can save time. It can summarize documents, improve writing, help analyze information and assist with technical problems.

Blocking every AI tool means losing those benefits.

It can also push employees toward other AI services that the company knows even less about.

A more practical starting point is visibility.

First understand which AI tools employees are actually using. Then decide which tools are acceptable, what information can be shared with them and what information should never leave company systems.

What Should Companies Do?

The first step is surprisingly simple: find out what is already happening.

Before writing a long AI policy, companies should understand which AI services their employees currently use and how common that usage is.

The next step is defining what should never be shared.

For one company, that might include customer lists and pricing. For another, it could be source code, contracts or financial information. Law firms may be particularly concerned about client documents and confidential case information. A software company may care more about source code and credentials.

Employees also need simple guidance.

"Do not share confidential information with AI" sounds clear until someone needs ChatGPT to summarize a 40-page document. Rules become much more useful when employees are given real examples of what is and is not acceptable.

Higher-risk organizations can then introduce DLP controls, alerts or restrictions for specific types of information.

How MonitUp Fits Into This Today

MonitUp already helps companies understand what happens on their Windows computers.

Through its employee monitoring software, MonitUp records application and website usage, working times and activity patterns, giving managers visibility into how company computers are being used.

This includes visibility into web-based AI tools. If employees are accessing services such as ChatGPT during the working day, organizations can understand when these services are being used and how they fit into overall computer activity.

There is an important limitation, however.

Seeing that someone used ChatGPT is not the same as detecting that sensitive data was sent to ChatGPT.

MonitUp should not currently be considered a content-aware DLP solution.

We believe the next interesting question is whether employee activity data can help companies identify unusual AI usage and potential data-risk signals without forcing administrators to manually review everything employees do.

It is an area we are actively exploring.

The Question Companies Need to Start Asking

For the last few years, companies have spent a lot of time asking whether employees should be allowed to use ChatGPT.

That question is becoming less useful.

Employees are already using AI tools.

The more important question is what happens when they use them.

Are employees only asking ChatGPT to improve an email?

Or are customer records, product prices, contracts, payment information, source code or other confidential data being submitted along with those prompts?

Most employees will not intentionally create a security problem. They will simply use whatever tool helps them get their work done.

For companies, the challenge is finding a balance between allowing employees to benefit from AI and maintaining visibility over where company information goes.

That is likely to become a much bigger part of employee monitoring and data security as AI becomes a normal part of everyday work.

Sources

Netskope: Cloud and Threat Report 2026

Microsoft Learn: Prevent data leakage to Shadow AI with Microsoft Purview

OWASP GenAI Security Project: Sensitive Information Disclosure