AI Act may push the HR industry towards technological regression

The recent entry into force of key provisions of the EU AI Act is transforming the operational and legal landscape of the HR industry. Systems used for CV screening, candidate assessment and promotion prediction have officially been classified as high-risk systems. Transparency requirements have already entered into force, while the obligation to audit data and prevent algorithmic discrimination will apply from December 2027. However, the new regulations in no way protect companies from candidates using AI in recruitment processes and may lead to technological regression in the HR industry. The solution lies in a well-designed architecture of processes that consciously combines the computing power of machines with human empathy and contextual understanding – so-called natural intelligence.

The illusion of oversight and the trap of automated bias

Under the AI Act requirements, employers must ensure effective human oversight of artificial intelligence systems. This is intended to reduce the risk of incorrect or unfair decisions made by algorithms during recruitment processes.

The influence of algorithms on a candidate’s career path often begins even before an application is submitted. Systems may decide who sees a particular job offer, search for potential candidates and create initial lists of people considered in the recruitment process. The obligations resulting from the AI Act do not end when an employee is hired. Artificial intelligence is increasingly used later as well, including for scheduling, task allocation, performance evaluation, bonus allocation and promotion decisions. It is worth emphasising, however, that not every AI system used in HR is considered high-risk – the key factor is whether its output influences a decision concerning an employee or candidate.

If, from a database of 1,000 candidates, an algorithm selects 50 people and an exhausted recruiter simply approves the generated shortlist, human oversight becomes a legal fiction. This means that the decision to exclude 950 people was made entirely by a machine, which will not comply with the law. Therefore, the new legislation is expected to require companies to pay greater attention to the quality of data used by systems, document their operation and ensure genuine human oversight of decisions made. It will no longer be sufficient for an HR employee to formally approve a decision generated by an algorithm – they will need to be able to verify and challenge it.

– Oversight of an algorithm cannot be reduced to the role of a mindless mouse operator. When a recruiter receives an extensive list of applications and, under time pressure, simply stamps approval on a machine’s selection, the risk of reproducing hidden biases increases dramatically – says Natalia Myskova, Managing Director of Smart Solutions HR. – Real control exists only when a system’s recommendation is treated as an initial stage of work and approached with significant caution. A person must have the time and competence to challenge selection criteria and verify why a particular profile was excluded from the process.

The new obligations affect the main efficiency indicator of HR departments – time to hire. Previous practices rewarded speed and enabled faster identification of suitable candidates without requiring a deeper understanding of the criteria used by the system. Today, the priority becomes understanding the software used in recruitment processes. The new law requires companies to implement tools that provide clear explanations of decisions, precisely indicating which parameters or keywords influenced the candidate’s assessment.

– The answer to business pressure is a hybrid selection model. While artificial intelligence can objectively automate the verification of hard skills, the assessment of cultural fit should be based on natural intelligence – says Myskova.

Legal asymmetry: what does the AI Act not protect employers from?

In the new reality, a significant challenge for HR processes is the asymmetry of the regulation itself. The AI Act is a mechanism protecting candidates from employers’ algorithms, but at the same time it does not regulate the use of artificial intelligence on the applicants’ side in any way.

– The new regulations work in one direction – the protective shield covers the candidate, but leaves the organisation vulnerable to a wave of synthetic, perfect applications generated by AI – notes Natalia Myskova. – Since recruitment has been classified as a high-risk area, the entire burden of auditing and transparency falls on the company. This asymmetry means that traditional CV assessment is becoming increasingly risky. An effective defence against mass application spam is a radical shift towards live competency verification, where natural intelligence once again becomes a filter that a candidate’s algorithm cannot manipulate.

Future skills: data and AI literacy

High penalties for violating the AI Act create a temptation in the industry to return to manual processes, especially in areas of soft HR (sentiment analysis, employee attrition prediction). However, rejecting automated analytics means losing competitiveness.

Therefore, the key is not abandoning technology, but understanding the new structure of responsibility. The employer acts as the entity implementing the technology and bears direct responsibility towards the candidate and the supervisory authority for any potential discrimination. The software provider is responsible only for the technical compliance of the system. Future agreements with technology partners must precisely regulate the employer’s right to audit and the mechanisms for storing the system’s operational history for the six months required by regulations.

The transformation of the HR industry does not mean the need to hire AI ethics specialists. The answer is a model in which support from external legal firms in the area of audits is combined with the development of internal, multidisciplinary teams.

– Instead of succumbing to legislative paralysis and returning to the era of analogue processes, organisations must acquire new competencies. HR departments do not need to hire full-time AI ethicists, but HR professionals must absolutely master data literacy skills – concludes Natalia Myskova. – They must understand data architecture well enough to ask technology providers challenging questions, interpret technological limitations and consciously decide when an algorithm must hand control back to a human.

In the long term, the HR software market will face standardisation and alignment with the “Compliance-by-Design” approach – creating systems that comply with regulations from the very moment they are designed. The winners of the new era will be those organisations that use artificial intelligence to eliminate repetitive processes, regaining time to build what lies at the foundation of HR – authentic, empathetic human relationships.