The Signal · Column 070

The EU Makes the Employer Own the AI Hiring Decision

Employment lawyers have started publishing the compliance checklists, which is usually the sign that a technology has stopped being interesting and started being real. The guidance for European employers is that deploying an AI recruitment tool in the EU triggers obligations under both the GDPR and the EU AI Act, and those duties attach to the company doing the hiring, not only to the company that wrote the software. You cannot sign your way out of them in a vendor contract.

Here is the part most hiring plans miss. Screening was the first place AI got pointed at because it looked like the safest place, a back office task with clean inputs and a tidy output. Under this framework it is the opposite. A rejection is a decision about a person, and a decision about a person needs a basis, a notice, a record, and somewhere a human who can stand behind it. The tooling moved the work out of the recruiter's inbox. It did not move the responsibility anywhere at all.

Automation moved the work. It did not move the accountability.

So the market reprices again, and in a direction almost nobody budgeted for. A company running thousands of applicants through a model now needs someone who can look at the resulting shortlist and explain, in plain language, why the top of it is the top of it. That is not a coordinator, and it is not a compliance officer bolted on after the fact. It is a recruiter with enough technical literacy to interrogate the system and enough seniority to put their name under the answer. That person was not on the org chart anyone drew when the tool was bought.

I expect the quiet failure mode to cost more than the loud one. Very few companies will be made an example of. Most will simply pay a slow tax instead: candidates who never learn why, hiring managers who stop trusting the pipeline, and a team that cannot reconstruct its own decisions when someone asks six months later. None of that shows up in the efficiency case for the software.

For hiring leaders, the practical read is simple. Before you switch on another screening layer, name the person who owns its output, and give them the standing and the hours to genuinely read what it produces. If no such person exists on your team, you have not bought a tool. You have bought an exposure.

Andrei, Founder

Planning a senior AI or data hire and want the market view first? We share it either way.

Send us the brief →