The Signal · Column 006

The AI Act's Hiring Rules Are Already Late

Andrei van Noordt · 20 July 2026 · 1 min read

Starting this month, the EU AI Act requires labels on chatbots and deepfakes. But the provisions that matter most to anyone building or buying hiring and screening tools, the rules on biometrics, migration, and employment-related AI, have been pushed back to December 2027.

Most readers will treat that as a reprieve. It is not. A regulation announced years in advance with a distant deadline is exactly the kind of signal that separates companies who build capability early from companies who scramble later. The gap between now and 2027 is not free time. It is the window in which competent teams get built, tested, and made permanent, while everyone else waits for the deadline to force their hand.

A deadline three years out is not permission to wait. It is a head start being handed to your competitors.

This is why I keep telling clients that compliance hiring is really judgment hiring in disguise. The people who understand how hiring algorithms actually behave, who can document a model's logic well enough to survive an audit, who know the difference between a defensible screening tool and a lawsuit waiting to happen, are not commodity compliance hires. They sit at the intersection of legal literacy and technical fluency, and there are very few of them in any given market.

The talent market has not caught up to this yet. Companies are still hiring generic AI governance staff, the kind who can recite the regulation but cannot read a model card. The candidates worth paying for are the ones who have already built or audited a real hiring or biometric system under scrutiny, because that experience cannot be taught in the eighteen months before enforcement lands.

For hiring leaders, the practical read is simple. Start building or acquiring that judgment now, while the market still underprices it, rather than in 2027 when every competitor is bidding for the same small pool at once.