Key Takeaways
- The EU AI Act's full "high-risk" compliance package for recruitment AI — bias audits, technical documentation, human oversight processes, conformity assessment — was due 2 August 2026. It has since been pushed to 2 December 2027 by Regulation (EU) 2026/1744, the "Digital Omnibus."
- That delay doesn't touch two rules already in force: the ban on AI inferring candidates' emotions, stress or personality traits (since 2 February 2025), and the duty to disclose when AI is involved in a hiring decision (since 2 August 2026).
- It doesn't touch ordinary anti-discrimination law either. Age discrimination under Directive 2000/78/EC is illegal whether a human or an algorithm makes the call — the AI Act's timeline is irrelevant to that.
- 39% of organisations already use AI somewhere in HR, but only 27% specifically in recruiting (SHRM, State of AI in HR 2026) — most companies still have real runway to get this right before 2027.
- Candidates already have enforceable rights today: advance notice, a route to human review, the right to refuse emotion/voice/body-language analysis, and access to (and deletion of) the data collected about them.
Since August 2026, articles warning employers that AI hiring tools now require "a qualified human in the loop" have circulated widely. Some of that is accurate. Some of it describes a deadline that, by the time those articles were published, had already been pushed back by more than a year.
For a company hiring in only one EU country, mixing those two things up is a paperwork problem. For a company running recruitment across Italy, Germany, the Netherlands and beyond, it's the difference between panicking over a 2026 deadline that no longer applies and missing two rules that were never delayed at all.
How we built this analysis
We're not an AI compliance auditor, and this isn't a substitute for legal advice. This is based on the text of Regulation (EU) 2024/1689 (the EU AI Act) and its amending Regulation (EU) 2026/1744 (the "Digital Omnibus"), Council Directive 2000/78/EC on equal treatment in employment, and SHRM's State of AI in HR 2026 report. Where a specific figure or date is disputed or still moving, we've flagged that rather than stated it as settled.
What's Already Enforced Today
Two obligations relevant to hiring have applied since before this article was written, and neither was touched by the 2026 delay described below.
- The emotion and personality inference ban (Article 5(1)(f), in force since 2 February 2025): no AI system may infer a candidate's emotions, stress level or personality traits from their face, voice or body language and turn that into a score. Hiring is explicitly covered as a workplace context.
- Transparency to candidates (Article 50, in force since 2 August 2026): if AI plays a role in evaluating someone, they have the right to be told.
- Both apply to any company hiring people in the EU, regardless of where the company itself is headquartered.
Not sure whether your current recruitment stack already breaks one of these two rules? We can check it before a candidate — or a regulator — does.
Talk to an ExpertWhat Just Got Postponed to 2027
The AI Act classifies recruitment AI — CV screening, candidate ranking, video-interview analysis, chatbots making pre-selection calls — as "high-risk" under Annex III. That classification carries a heavier compliance package: risk-management systems, technical documentation, logging, bias testing, human oversight procedures, conformity assessment, registration in an EU database. That package was originally due 2 August 2026.
Regulation (EU) 2026/1744 — the "Digital Omnibus" — postponed it to 2 December 2027 for stand-alone high-risk systems like most recruitment tools (2 August 2028 for high-risk AI embedded in other regulated products). This is a confirmed legislative change, not a proposal still being negotiated.
Emotion & Personality Inference Ban
AI can't score how a candidate feels, sounds or behaves. No recruitment exception.
Transparency to Candidates
Deployers must disclose when someone is interacting with, or being assessed by, AI.
High-Risk Compliance Package
Bias audits, technical documentation, human oversight processes and conformity assessment for recruitment AI (Annex III).
Ordinary Anti-Discrimination Law
Age discrimination under Directive 2000/78/EC applies regardless of who — or what — made the decision.
A later deadline for the AI Act's paperwork is not a grace period for age discrimination. That law was never running on the AI Act's clock to begin with.
Where This Leaves Recruitment AI in Practice
Legally, employers have more runway on the formal high-risk compliance package than the original timeline suggested. Three things don't wait for 2027, though:
- Any tool that scores emotion, tone or personality needs that feature switched off now — not audited later.
- Candidates need to be told, in plain terms, that AI is involved in evaluating them.
- Every AI-influenced outcome remains subject to ordinary discrimination law. If a tool disproportionately screens out older candidates, "the vendor built it that way" is not a defence.
This is where it matters whether your Employer of Record or HR outsourcing partner is actually tracking EU-level AI and employment law, not just running payroll on autopilot.
What Candidates Can Still Demand From You
Independently of where your own compliance project stands, candidates already have real, enforceable rights in this process:
- Advance notice that AI is involved, and roughly how.
- A human being who reviews — and can override — an AI-influenced decision, on request.
- The right to decline having their facial expressions, voice tone or body language analysed during a video interview.
- Access to the data collected about them, and its deletion once the process ends.
Readiness Checklist
- Audit every tool in your hiring stack for any emotion-, stress- or personality-scoring feature — disable it now, not by 2027.
- Add a line to job ads or interview invites disclosing that AI is used, and how.
- Confirm a named, qualified person reviews — and can overturn — every AI-influenced hiring decision today.
- Start basic bias monitoring and documentation now; building the habit ahead of 2027 is cheaper than building it under deadline pressure.
- Ask your HR/AI vendor directly whether their tool's outputs correlate with candidate age, and get the answer in writing.
- Apply the same standard in every EU country you hire in — Italy included, on top of its own existing labour law obligations.
Not sure where your hiring process stands on this checklist? Let's map it together.
Talk to an ExpertFAQ
Is AI banned from making hiring decisions in the EU?
Not outright. But AI can't score a candidate's emotions or personality, and the full "high-risk" compliance requirements for recruitment AI aren't enforceable until 2 December 2027.
What EU AI Act rules on hiring AI are already in force today?
The ban on emotion- and personality-inference AI (since 2 February 2025) and the duty to disclose AI's role to candidates (since 2 August 2026).
What was delayed, and until when?
The Annex III high-risk compliance package for recruitment AI — bias audits, technical documentation, human oversight processes, conformity assessment — postponed from 2 August 2026 to 2 December 2027 by Regulation (EU) 2026/1744, the "Digital Omnibus."
Does the delay mean age discrimination via AI is legal until 2027?
No. Directive 2000/78/EC's ban on age discrimination in employment applies regardless of the AI Act's timeline, or of whether a human or an algorithm made the decision.
Does this apply to companies outside the EU?
Yes — if the hiring decision affects someone in the EU, it applies regardless of where the hiring company is headquartered.
Hiring in the EU and using any AI in the process? We'll tell you exactly what's binding today, what's coming in 2027, and what to fix first.
Talk to an Expert
