Recruitment edition

The Dutch Data Protection Authority (Autoriteit Persoonsgegevens) has fined Uber more than 800 million euros. The reason: Uber deactivated drivers in a fully automated way, with no human involved. It's one of the largest fines ever issued under the privacy law GDPR for automated decision-making. This isn't about taxis. It's about the principle that a computer may not make a serious decision about someone's job without human involvement. That exact principle applies to your work too. Every week you make decisions about people: who you invite, who you reject, who you pass on to the hiring manager. With this, the regulator is telling every organisation that uses AI in staffing or contract decisions: check your human review. That's a direct signal for recruitment teams working with AI screening, match scores, or automatic rejections. Not sometime in 2027, but now, because this fine rests on the GDPR, and that law has applied for years. Notice what the case does and doesn't say. Uber wasn't fined because software was part of the process. It was fined because no human was involved anymore in a decision with major consequences. So a tool that ranks candidates isn't banned. A knock-out question that automatically removes someone from your funnel, with no one looking at it, is more sensitive. This week, walk through your own process and write down, for each step, who makes the decision: you or the system. In practice, the gap isn't in your final decision, it's in the steps before it. A candidate fills in a form, a filter drops them for a knock-out question or a minimum score, and they never reach your screen. Formally, you decided. In practice, the system decided. A ranking works the same way: the list itself isn't a decision, but whoever is at the bottom is barely ever read. Human review only counts if that human also sees the people the system filtered out. The line is trickier than it sounds. If every rejection had to be checked by hand, you couldn't process thousands of applications, and candidates would then wait weeks for an answer. The workable middle ground: let AI sort and prepare, but keep the rejection with a human, and log per vacancy who was excluded and on what grounds. That log is also your defence. A regulator doesn't ask if you had good intentions. It asks if you can show that a human looked.

Source →