← Insights

Regulation

The EU delayed its rules for AI in recruitment. We’re not waiting.

Brussels pushed the high-risk deadline to the end of 2027. Our standard for recruitment AI hasn’t moved.

Arsénio Ferraz — CEO & Co-Founder, Venora · 20 August 2026 · 4 min read

Europe’s rules for “high-risk” AI, including the AI that helps decide who gets hired, were due to come into force this month. Brussels has pushed them back to the end of 2027.

We’re not pushing anything back.

What actually changed

The EU AI Act classifies AI systems used in employment, recruitment, candidate screening and selection, as high-risk. In practice, that classification turns good practice into legal obligation: transparency towards candidates, human oversight of outcomes, monitoring for bias, technical documentation and logging.

Those obligations were scheduled to apply from 2 August 2026. In June 2026, the EU approved a “simplification” package (the so-called Digital Omnibus) that postponed the high-risk deadlines by roughly sixteen months, to December 2027.

I understand the argument for the delay: give companies time to prepare, don’t strangle European innovation in paperwork. Some of that is fair.

But there’s a trap hiding in it: confusing a legal deadline with a quality standard.

A hiring decision touches someone’s life

Recruitment isn’t a low-stakes use case for AI. On the other side of every application is a person’s livelihood. A system that ranks or rejects candidates on its own, without explaining why and without anyone supervising it, is a bad system in 2026, in 2027, and after. The Brussels calendar doesn’t change that.

That’s why the delay changes nothing about how Venora is built:

  • Explainability isn’t a compliance feature. When Venora suggests that a candidate fits a role, it shows the reasoning. A recruiter who can’t see why can’t meaningfully disagree, and being able to disagree is the whole point.
  • The final decision always sits with a person. Venora assists; it doesn’t decide. That was our product philosophy before it was anyone’s regulation.
  • Bias is monitored, not assumed away. AI doesn’t remove human prejudice by default; it can inherit and amplify it. Treating that risk seriously is table stakes, deadline or no deadline.

For us, the AI Act was never the reason to build this way. It was confirmation that the direction was right.

Three questions to ask any recruitment AI vendor

If you run an SME and you’re considering AI in your hiring, this year, not in 2027, the delay makes your own diligence more important, not less. Three questions separate serious tools from shortcuts (for the longer checklist, see our honest guide to AI in recruitment for SMEs):

  1. Does the system explain why it recommended each candidate?
  2. Does the final decision always rest with a person?
  3. Do you know what data it uses to learn?

If any answer sounds like “that’s not mandatory anymore”, be doubly suspicious. A vendor who treats candidate fairness as a regulatory burden will treat your hiring the same way.

The deadline moved. The questions didn’t.

Regulation will catch up eventually; it always does. But the companies that will be comfortable in December 2027 are the ones choosing transparent, supervised, explainable AI today, not because they have to, but because it produces better hires and treats candidates like people.

That’s the standard we hold Venora to. No extension requested.