Like it or not, AI is already part of your workplace. Here’s how to keep it under control

If your job involves emails, reports, spreadsheets, or meeting notes, there is a good chance artificial intelligence is already helping out. But in most cases, it feels like AI didn’t knock before entering the workplace in Czechia. It just sat down and opened a laptop, leaving many companies in an awkward position. It’s blatantly clear now that AI use is globally widespread, but rising expectations and lagging usage policies are causing costly issues.

David Kaláb, vice president of data management at the Prague-based tech innovator Adastra, reveals how organizations can prevent the massive financial loss and ruined reputations that result from a lack of AI oversight. Much of it comes down to a cohesive company strategy, bespoke AI tools, and a focus on data security that frees employees to safely innovate. 

Why AI at work has become a governance problem

Workers en masse are making their own AI rules that frequently violate company policies and put sensitive company data at risk. A 2025 cybersecurity analysis revealed that 68 percent of organizations have experienced data leaks linked to the use of AI tools, yet only 23 percent have formal security policies in place to address these risks. 

Nearly half of employees who use AI at work admit to using it “inappropriately,” according to a major global study by Melbourne Business School. The AI anarchy in the workplace has leaders scrambling for policies and protections to prevent BYOA, Bring Your Own AI. 

Most employees are not trying to break rules. They turn to AI to save time, handle repetitive tasks, or keep up with growing demands. The problem is that public AI tools operate outside company security systems, allowing sensitive data to drift beyond control. 

To understand where companies are losing control and how they can regain it, Kaláb provides more in-depth insights.

More than half of employees admit they use ChatGPT at work, mostly without oversight. Why are companies letting this happen?

It’s a continuation of something we’ve seen for years with what is often referred to as shadow IT. When people need tools that make their work easier, they simply use whatever is available. ChatGPT became widespread long before most companies built secure, internal alternatives. So employees reach for the public tool, not out of rebellion, but because they want to get their job done faster. But that creates risk, because they often don’t realize they’re using it on sensitive inputs or in ways the company can’t control.

What are the biggest dangers when employees use public AI tools without guidance?

The first risk is data sensitivity. If someone uploads HR information, medical details, internal customer data, or anything that falls under the EU’s high-risk categories, the consequences can be serious.

The second danger is misunderstanding how these tools work. Employees sometimes assume they operate inside the company’s secure environment — but the public versions don’t. And then there’s the simple fact that generative AI doesn’t “know” the company context, so it may produce outputs that are simply wrong.

But by now most people realize that AI isn’t always accurate, right? Do they really rely on AI without verifying a chatbot’s information?

You only need to check out the latest business news headlines to answer that question. Last year a major global firm was called out for submitting government reports to Canada and Australia that included fake citations, nonexistent experts, and errors caused by AI ‘hallucinations.’ This is not only a massive reputation issue, but misused AI can result in substantial financial loss both for the firm distributing inaccurate information and for clients who rely on it.  

Companies are racing to use AI agents for a wide range of services, internally and externally. Will this boost productivity?

Before we get to productivity, leaders must recognize AI doesn’t understand a company’s actual processes. An AI tool vendor may promise the world, but they don’t know how your business operates day-to-day.

We joked recently in our podcast about an “agent zoo”: imagine everyone in the company building their own small agent for their own piece of work. One agent orders milk, another one does too, a third one checks the fridge. Suddenly the company has a mountain of milk and no one knows which automation did what. That’s what happens when AI is deployed without governance or strategy.

So what should companies do before letting employees use AI tools broadly?

First, they need clear governance: who can use which tool, on what data, and under what conditions.

Second, they need secure AI tools designed for the company’s unique needs that respect internal data access rights.

And third, they need education. Employees must understand what the tool can and cannot do, how data sensitivity works, and where automation is appropriate.

How did you handle this challenge at your own company?

At Adastra, we faced the same challenge as many companies: people wanted to use AI right away, and they were already experimenting with public tools. Instead of trying to block everything, we built a safe alternative, our internal AI assistant, Agate.

Agate gives every Adastran access to advanced models like GPT-5.5 and Claude, but inside a secure environment with clear data-handling rules. That means our people get the power of AI without risking client data or creating rogue workflows no one can manage.

What’s important is that Agate isn’t just a chatbot. It’s becoming a central platform for how we work with AI. We use Agate to support automation, coding help, summaries, structured documents, and decision support, all in one place. Over 300 colleagues use it already.

If you had to give one piece of advice to leaders worried about misuse of ChatGPT at work, what would it be?

Don’t try to block AI; guide it. Employees are already using these tools, whether you approve it or not. The only sustainable approach is to give them safe, governed alternatives and a clear framework. And above all, build your AI strategy on top of high-quality, well-connected data. If that foundation isn’t there, no tool, not even the smartest agent, will help you long-term.