The EU AI Act delay for hiring covers exactly one chapter: the high-risk duties for recruitment and employment AI now apply from 02-12-2027 instead of 02-08-2026, under Regulation (EU) 2026/1744, the Digital Omnibus on AI. Everything else is live, and the Commission's AI Office and national authorities have been enforcing the Article 50 transparency rules since 02-08-2026. If you screen with AI, run a careers-page chatbot or score candidate quizzes, that split matters. Here is what the Omnibus changed, what an EU employer must have in place today and what to build in the 13 months to December 2027. General information, not legal advice.
- Only the high-risk chapter moved: Annex III recruitment and employment AI now applies from 02-12-2027, a 16-month extension under Regulation (EU) 2026/1744, in force since 27-07-2026.
- Article 50 transparency has applied since 02-08-2026 and the Commission announced enforcement from that day; breaches carry fines of up to EUR 15 million or 3 per cent of global turnover.
- AI literacy (Article 4) was softened, not scrapped: you must take measures supporting staff AI literacy, not guarantee a proficiency level.
- Explained scores, a human decision and a record of the criteria are GDPR Article 22 duties today, not something to park until December 2027.
What did the Digital Omnibus actually change for hiring?
One date. Regulation (EU) 2026/1744 was published in the Official Journal on 24-07-2026 and entered into force three days later, on 27-07-2026 (Cloud Security Alliance, 01-08-2026). It moves the high-risk obligations for standalone Annex III systems, the category that covers recruitment, selection and worker management, from 02-08-2026 to 02-12-2027. High-risk AI embedded in regulated products under Annex I moves to 02-08-2028. The Council gave final approval on 29-06-2026, after a provisional deal on 06-05-2026 (Gibson Dunn, 27-05-2026).
Everything else stayed put. The prohibited practices and the AI literacy duty have applied since 02-02-2025. Article 50 transparency landed on schedule on 02-08-2026. So "the AI Act was delayed" is the wrong summary: the heavy compliance chapter for hiring tools got a 16-month extension, and the lighter duties that touch most employers are in force. The full walk-through of the Act is in our EU AI Act hiring guide. This post sticks to what moved and what did not.
Which transparency rules have been enforced since August 2026?
On 31-07-2026 the Commission announced that from 02-08-2026 the AI Office, working with national authorities, would start enforcing the transparency rules: chatbots and other interactive AI systems must tell people they are dealing with AI, deepfakes must be labelled and AI-generated content must carry machine-readable marks. More than 180 organisations had signed the Code of Practice on transparency by then (European Commission, 31-07-2026). Breaches of Article 50 carry fines of up to EUR 15 million or 3 per cent of global annual turnover (Jones Walker, 16-07-2026), with separate caps for SMEs and small mid-caps under the Omnibus.
Read the article before you panic. Article 50(1) is a duty on the provider of a system built to interact directly with people. A careers-page chatbot or a voice screening agent must say it is AI unless that is obvious. The deployer duties in 50(3) and 50(4) cover emotion recognition, biometric categorisation and deepfakes. A scored quiz is not caught by Article 50 on its own. You still tell candidates about it, but that duty comes from GDPR, and from common sense.

What does GDPR already require when AI scores candidates?
This is the part employers keep parking until 2027, and it is live now. GDPR Article 22 restricts decisions based solely on automated processing that produce legal or similarly significant effects, and a job rejection is one of them. In practice: a human makes the decision, the candidate is told how the score was produced, and they can contest it. Articles 13 and 14 mean your candidate notice mentions the automated scoring before anyone takes the test. GDPR for recruiters covers retention periods and access requests.
Two Omnibus footnotes matter here. The final text restored a stricter standard for processing special categories of personal data for bias detection and correction (Hunton Andrews Kurth, 28-07-2026), so if you test a screening tool for bias against protected groups, check the legal basis before you collect that data. And the separate Digital Omnibus on data and GDPR is still going through the legislative process: the Council agreed its negotiating position on 21-10-2026 and kept the current definition of personal data (Council of the EU, 21-10-2026), with talks with Parliament still to come. Nothing in GDPR has changed yet, whatever a vendor slide says.
The numbers here are illustrative, drawn from Qwiza demo campaigns and pilot targets, and the names are invented. Marta runs operations at a courier company with 35 staff and hires around 20 drivers a year through quiz ads. She has been told the AI Act is "delayed" and wants to know what that means for her.
In one afternoon she does four things. She adds two sentences to the quiz landing page saying answers are scored automatically and a person reviews every application. She exports the question set with a one-line note on what each question measures. She names herself as the person who signs off every rejection, and she asks her vendor in writing whether the tool will fall under Annex III and what documentation they will provide before 02-12-2027.
In the demo scenario, a campaign of 120 quiz completions produces about 30 flagged for a closer look on timing or pasted text. None are rejected on the flag. Marta reads each one and rejects 9 with a note. That log is her GDPR Article 22 record today and the start of her Article 26 file for 2027.
What has to be ready before 02-12-2027?
From 02-12-2027 an employer using an Annex III recruitment system becomes a deployer of high-risk AI. Three duties attach: human oversight under Article 26, the right to an explanation under Article 86, and Article 26(7), which requires you to inform workers' representatives and the affected workers before the system goes into service. The Omnibus did not touch Article 26(7); it moves with the chapter (EU AI Act Explorer, Article 26). Most documentation sits with the provider, and the Omnibus extended some simplified documentation measures to medium-sized companies (Hunton Andrews Kurth, 28-07-2026).
Thirteen months is plenty if you start with four things:
- Vendor question: ask in writing whether the tool will fall under Annex III and what documentation they will hand you before December 2027.
- Criteria record: write down what each screening question measures and why it relates to the job.
- Named decider: one person signs off every rejection, and the log shows it.
- Candidate notice: a plain paragraph in the application flow saying an automated score exists and a human decides.
How does a compliant AI screening setup look in practice?
Take a quiz ad. The ad opens straight into a two-minute quiz, no CV, no login. Your ad copy and landing page say answers are scored automatically and a person reviews every application. You wrote and graded the answers, so the criteria record already exists. Graded answers plus time per question and completion become one score with a written reason, which is the explanation Article 22 wants today and Article 86 will want from December 2027. A recruiter reads the score, adds a pipeline note and decides. The pipeline keeps the record.
Integrity signals are where discipline matters. Qwiza flags per-question timing, tab switching and pasted text so a recruiter can take a closer look. That is a prompt, not a verdict. Turn a flag into a rejection rule and you have built the automated decision the law is aimed at. The quiz screening guide covers what to score and what never to automate. Pick tools that already explain themselves and December 2027 becomes a paperwork check, not a migration.
Screening that explains itself
Qwiza turns a job description into a quiz ad, scores every answer with a written reason and leaves the decision to a person. First campaign free.
Try the live demoFrequently asked questions
Was the EU AI Act delayed?
Only one chapter. The Digital Omnibus on AI, Regulation (EU) 2026/1744, in force since 27-07-2026, moved the high-risk obligations for standalone Annex III systems, including recruitment and worker management, from 02-08-2026 to 02-12-2027, and product-embedded Annex I systems to 02-08-2028 (European Commission, 27-07-2026). The prohibited practices and the AI literacy duty have applied since 02-02-2025, and the Article 50 transparency rules since 02-08-2026. If someone tells you the whole Act is on hold, they are reading a headline, not the regulation.
Do I have to tell candidates that an AI scored their quiz?
Yes, but the duty comes mainly from GDPR rather than Article 50 of the AI Act. Article 50(1) obliges providers of systems that interact directly with people, such as chatbots and voice agents, to make that clear. A scored quiz is not a chatbot. GDPR Articles 13 and 14 still require you to tell candidates about automated scoring before they take it, and Article 22 requires a human decision and an explanation they can contest. From 02-12-2027, Article 26(7) of the AI Act adds a duty to inform workers' representatives before a high-risk system goes into service.
Is AI literacy training still required for hiring teams?
Yes. Article 4 has applied since 02-02-2025, and the Omnibus softened it rather than removing it. The final text asks providers and deployers to take measures supporting AI literacy among staff, instead of guaranteeing a sufficient level for each person (Gibson Dunn, 27-05-2026, confirmed in the final regulation). For a small employer that means a short session on what the screening tool does, what its score means, where it can be wrong and who makes the final call. Keep a note of who attended. It also happens to be the training that makes the tool useful.


