If you apply to a mid-sized company, it is increasingly likely that the first “reader” of your resume is not a person. An Applicant Tracking System (ATS) parses the text, matches it against keywords, sometimes assigns a score, and only some of the applications ever reach a human recruiter. This article explains what European law says about these systems, what changed recently (and almost no online content has caught up yet), and what you can actually do as a candidate.
What the AI Act classifies as “high risk” in recruitment
Regulation (EU) 2024/1689 on artificial intelligence (the AI Act) does not treat all AI systems the same way. It sorts them by level of risk, and Annex III lists the uses considered “high risk” because they affect people’s fundamental rights.
Point 4 of Annex III explicitly covers employment, workers management and access to self-employment, and includes systems used to:
- place targeted job advertisements;
- analyse and filter job applications;
- evaluate candidates.
In practice: automated resume parsing with ranking, predictive candidate-to-role matching, scores assigned by software, video interviews with language analysis. All of this falls into the most closely supervised category of the regulation, alongside areas such as critical infrastructure and the administration of justice.
The postponement almost no article has picked up yet
Here is the part that makes a lot of content published until a few weeks ago out of date. Regulation (EU) 2026/1744, the so-called “Digital Omnibus”, was published in the Official Journal of the European Union on 24 July 2026 and entered into force on 27 July. Among its changes, it postponed the application of the obligations for Annex III high-risk systems from 2 August 2026 to 2 December 2027.
This means AI systems used in recruitment are still classified as high risk (the classification has not changed), but companies have more time before they must meet the specific obligations for that category: risk management, traceability, technical documentation, accuracy, human oversight.
What is mandatory right away, regardless of the postponement. The transparency obligations in Article 50 of the AI Act apply from 2 August 2026 anyway: if you interact with an AI system (for example an automated interview or a recruiting chatbot), the company must tell you that you are not talking to a person. The postponement concerns the heavier technical and organisational requirements for high-risk systems, not the duty to tell you that an algorithm is involved.
The legal tool that already exists, AI Act or not
Many articles on this topic make the same mistake: treating the AI Act as if it were the only protection available. It isn’t. The GDPR has applied since 2018 and, in Article 22, gives you the right not to be subject to a decision based solely on automated processing, including profiling, when that decision produces legal effects concerning you or similarly significantly affects you. Being rejected in a hiring process falls within that definition.
Recital 71 of the GDPR is even more explicit, and gives as an example precisely “e-recruiting practices without any human intervention”. This is not a stretched interpretation: the European legislator had exactly this scenario in mind back in 2016.
When Article 22 applies, the company must guarantee you at least three things: the right to obtain human intervention, the right to express your point of view on the decision, and the right to contest it. An ATS that automatically rejects a resume with no possibility of human review, when that decision significantly affects you, is exactly the case this rule is meant to regulate, whether or not the AI Act’s specific obligations have been postponed.
The closest Italian precedent
In Italy there is not yet a public sanction from the data protection authority (the Garante per la Protezione dei Dati Personali) on a resume screening system used in hiring. The most concrete precedent concerns the algorithmic management of workers, and it is instructive for understanding what the authority considers unacceptable.
In 2021 the Garante fined Foodinho (Glovo group) €2.6 million over the algorithmic management of its riders. In November 2024 it came back to the same company with a €5 million fine: among the issues raised was the automatic deactivation of accounts through a standard message that did not mention the possibility of contesting the decision. It was not a sanction about recruitment but about the automated management of workers, and the core point, transparency plus the right to contest, is the same principle as GDPR Article 22 applied to another work context. It is the clearest signal of how the Italian authority reacts when an algorithm decides without leaving a way out to the person affected.
What you can actually do as a candidate
Ask, directly. During an interview you can ask whether resumes are screened by an automated system. It is not an out-of-place question: it is exactly the information Article 50 of the AI Act requires to be provided.
Write your resume with its first reader in mind. If the first filter is an ATS, formatting rules matter as much as content: the guide on how to write an effective resume explains what these systems actually read and what they ignore.
Check the privacy notice attached to the job ad. If a company uses automated profiling in hiring, the notice must mention it. Its absence does not mean the system isn’t used (often it only means the notice is out of date), but it is a clue.
Know that you can ask for human review. If you receive an automatically generated rejection and suspect no person ever looked at your application, you can write to the company citing GDPR Article 22 and request the human intervention the rule provides for. It doesn’t guarantee a different answer, but it turns the request from “please reconsider” into “this is a right I am exercising”.
Know where to turn. If a company refuses to explain or to provide the review the law requires, you can lodge a complaint with your national data protection authority (in Italy, the Garante).
None of these actions guarantees a different outcome in the hiring process. They do change your negotiating position with a system that would otherwise remain a black box.
My view: the filter isn’t the problem, the opacity is
I don’t think an automated resume filter is a problem in itself. With hundreds of applications for a single posting, a first filter is needed, and rejecting it on principle is not realistic. The problem isn’t the filter, it’s the opacity. A company that uses these tools and says so clearly, leaving a way to ask for a review, is doing exactly what the law asks and what I consider the right thing to do. A company that uses them and denies it, or has no review process at all, is treating your application as data to discard faster, not as a request from a person.
Frequently asked questions
Can a company reject me just because an algorithm decided so? If that decision significantly affects you, and a rejection in a hiring process does, GDPR Article 22 gives you the right to request human intervention. It does not give you the right to prevent the use of an automated system altogether.
Does the 2026 postponement mean the AI Act no longer applies to recruitment? No. It means the specific technical and organisational obligations for Annex III are postponed to 2 December 2027. The transparency obligations in Article 50 still apply from 2 August 2026, and the GDPR applies regardless of the AI Act’s timeline.
How do I know whether a company uses these systems? By asking directly during the process, or by checking the privacy notice linked to the job ad.
Is it worth bringing up in an interview? Asking how the selection process works is a legitimate question and, asked in an informative rather than accusatory tone, it is generally read as interest in the organisation, not as distrust.
Sources
- Regulation (EU) 2024/1689 (AI Act): Annex III, point 4, high-risk classification of AI systems for recruitment and workers management.
- Regulation (EU) 2026/1744 (“Digital Omnibus”): postponement to 2 December 2027 of the obligations for Annex III high-risk systems.
- Regulation (EU) 2016/679 (GDPR): Article 22 and Recital 71, on automated decision-making.
- Italian Data Protection Authority (Garante): sanctions on the algorithmic management of riders (Foodinho/Glovo, 2021 and 2024).


