Your IT team has told you not to put employee data in an AI tool. The reason rarely comes with the no. I asked Tobias Lipsky, a strategy consultant in Vienna, what IT is weighing when it says that. This page turns the conversation into four questions to bring to the meeting. None of it is legal advice.

Most of IT's worries aren't new, he says. Who can access HR data, where it moves and what gets done with it were IT's questions long before AI. The only new party is the company behind the tool. So the meeting goes better when you know the questions IT already asks of any system that touches HR data.

IT will start with its own question, which is what you want to get done. Tobias says it's the answer he wishes people brought with them, because without a clear end result IT can build something that works and still isn't what you needed.

Ask this before you go shopping, because IT may point you to a tool it has already approved. Tobias's warning is about the alternative. A private paid account can mean you've agreed to let the provider save and use your data, and putting company data into it breaks company rules. He also says the exposure starts when you upload a file, before you press send.

Say what you're trying to accomplish and what's in the file. Salaries, equity grants, ownership percentages, merger and acquisition data and anything that identifies a person all call for different guardrails. Tobias says the more sensitive the data, the harder the guardrails IT will want.

Tobias uses a house. Your tools are the walls, how they connect and what you use them for are the rooms, and where the data sits is the foundation. Data residency asks which country the server is in. Sovereignty asks who owns and runs it, because a server in your own country can belong to a company outside it. His test is a kill switch. If somebody turned the tool off tomorrow, could you keep operating?

If IT blocks an AI tool while your files already sit with the same provider, he suggests asking what changes when you add the AI. That assumes the guardrails between departments are in place.

Pay teams already work with employee numbers instead of names, and Tobias says the AI can do the same. It works on the number, and you hold a separate table that matches numbers back to people. What matters is where that matching happens. Inside your company and out of sight of the next department, he says you're fine. Inside the AI tool and outside your company's reach, he calls it a bit of an issue.

He also wants you asking what stops a colleague in another department from prompting the AI and learning the timing and size of the next comp cycle. His answer is to carry your existing access rules into the AI rather than invent new ones.

IT will often push the decision up. Tobias says it frequently ends with the head of legal, who negotiated the software contracts, because a single sentence in a contract can rule out your most sensitive data.

Tobias says people already experiment by describing the same problem in a neutral setting. Someone in shipping would describe it as a train company, take the answers back and translate them to shipping. He says the same approach works with the AI your company already provides. It's the same idea as the made up company in the Claude skills guide.

The full conversation is on Spotify as Stop being scared of your IT team.