Skip to main content

Your job interview may involve AI. Brussels moved the deadline, not the rules

Ilustrační obrázek
A company in Brno screens 300 job applications with a chatbot that ranks candidates. Elsewhere, an applicant never learns that a machine filtered her out, and an HR manager assumes nothing applies to her yet. Both assumptions are wrong in different ways: the EU moved the deadline for high-risk hiring tools to 2 December 2027, while the ban on prohibited practices, the transparency duties and the AI literacy rule are binding now.

On 6 November 2024, The National Law Review published a written analysis of the EU AI Act’s business implications. It predates the 2026 simplification, so its compliance checklist is best read alongside the amended timetable. Read together with the amendment, its central point remains useful: dates have shifted, but the duties have not disappeared.

The amendment changed dates, not duties

The change comes from the Digital Omnibus on AI, Regulation (EU) 2026/1744, adopted on 8 July 2026, published on 24 July 2026 and in force from 27 July 2026. Recital 40 moves the Annex III stand-alone high-risk use cases — employment, education, credit, essential services and certain biometrics — to 2 December 2027.

Annex III covers AI used in recruitment, selection, promotion and workforce management, which is where most ordinary office AI lands once it touches people.

The obligations themselves stayed in the text. A tool that ranks CVs still has to be classified, registered in the EU database, documented, kept under human oversight, and meet the Act’s risk-management, data-governance, record-keeping, transparency, accuracy, robustness and cybersecurity requirements. What organisations gained is roughly sixteen months of preparation time, not an exemption. The amendment does not postpone the Article 5 prohibitions, which have applied since 2 February 2025, or the AI literacy duty, which has applied since the same date.

Four steps the analysis lists

The National Law Review analysis sets out a four-part sequence for organisations preparing for the Act.

  1. Carry out an impact assessment. That means building an AI inventory and asking which systems are used, developed or placed on the market; whether they qualify as AI systems under the Act; what the organisation’s role is — provider, deployer, importer, distributor or product manufacturer; and which risk category applies: prohibited practice, high-risk system or transparency risk.
  2. Implement tailored compliance mechanisms. The design should fit the organisation’s role and applicable risk category, as well as its size, culture and overall approach to managing compliance risks.
  3. Monitor the regulatory landscape. Guidelines, codes of conduct and implementing acts will further define the framework. The analysis notes that these documents matter for understanding detailed requirements and improving compliance, including for general-purpose AI models.
  4. Engage in policy dialogue. Businesses, particularly those dealing with high-risk systems and GPAI models, should consider taking part in European Commission consultations and engaging with regulators and market surveillance authorities to understand enforcement strategies and design a pragmatic approach.

ai-jarvis.eu’s own HR reading of the Act, not a conclusion of the National Law Review analysis: two practical additions matter for hiring teams. A manager pasting CVs into a free chatbot creates a “shadow AI” governance gap, and bias testing should look at outcomes as well as prompts. Where a fundamental rights impact assessment is required, it should be coordinated with an existing data protection impact assessment rather than duplicated.

The amended timetable at a glance

MilestoneDateScope
Prohibited practices, Article 52 February 2025Social scoring, manipulative techniques, emotion recognition in workplaces and schools
Transparency and labelling, Article 502 August 2026Disclosure that a person is talking to an AI; labelling of synthetic content
High-risk Annex III systems2 December 2027Recruitment, candidate evaluation, worker management
AI inside regulated products, Annex I2 August 2028Medical devices, machinery and similar

Penalties sit in two bands. Prohibited practices carry up to €35 million or 7% of worldwide annual turnover, whichever is higher. Breaches of other obligations, transparency included, carry up to €15 million or 3%. The European AI Office, established in January 2024, is central to enforcement, particularly for general-purpose models; it can evaluate models, request data from providers and enforce corrective measures.

What a candidate can ask

Transparency is the part most people will meet first. If an AI system interacts with a job applicant, the AI Act’s transparency duties require that the person be informed. Separately, automated decision-making triggers information and safeguards under GDPR Article 22. An employer needs a lawful basis for processing, and that basis is not automatically consent.

A rejected candidate in the EU can ask whether an automated system was involved in assessing her application, and on what basis. She cannot demand the model’s source code, and the reply may be limited where trade secrets apply. Under the amended schedule, the fuller set of high-risk documentation duties for that same system arrives in December 2027.

Contracts decide who answers

Most European employers buy screening software rather than build it. Under the Act, the vendor is usually the provider and the employer the deployer, and the deployer carries duties that cannot be outsourced by signature. Human oversight, monitoring and record-keeping stay with the company that uses the tool.

In practice, role allocation is where mid-sized firms often slip. A tool bought on a departmental credit card, used by three recruiters and never added to the vendor register, is invisible to legal until a complaint arrives. The Commission’s AI Act pages remain the reference text for the current timetable.

Does the December 2027 date mean my employer can use AI in hiring without limits until then?

No. The Article 5 bans have applied since February 2025, the transparency rules since August 2026, and the AI literacy duty since February 2025. The GDPR applies to automated decisions about applicants regardless of the AI Act timetable. What was postponed is the main body of Annex III high-risk obligations.

Is a chatbot that ranks CVs automatically high-risk?

A system intended to filter applications or evaluate candidates falls under Annex III point 4, and that generally makes it high-risk. Classification depends on the intended purpose, which is why the first compliance step is an inventory rather than a legal opinion.

What does AI literacy actually require?

Providers and deployers must take measures so that staff dealing with AI systems have sufficient understanding of them, taking their technical knowledge and context into account. For an HR team that means practical training on the tools they use, not a general lecture on machine learning.

The National Law Review analysis, published on 6 November 2024, remains a useful checklist for employers. The most concrete date for HR teams now is 2 December 2027 for Annex III systems. Until then, Article 5 prohibitions, the transparency duties in force since August 2026 and the AI literacy requirement stay enforceable, with maximum penalties of €15 million or 3% of worldwide annual turnover for breaches of the general obligations.

Discussion

No comments yet — be the first to share your thoughts.
X

Don't miss out!

Subscribe for the latest news and updates.