AI in HR: Opportunities and Limits of Using Artificial Intelligence in Employment Relationships

Artificial intelligence is gradually becoming a routine part of human resources management. It is used in recruiting employees, assessing CVs, evaluating work performance and setting remuneration.

In employment relationships, the question is not only whether AI can technically handle a given task. What also matters is whether its use is legally permissible, what data the system processes, whether its operation leads to discrimination, and whether the employer is genuinely making its own decision or merely formally confirming a result produced by an algorithm.

One of the most common uses of AI in HR is the automated processing of CVs and job applications. The system can filter and compare candidates and produce a ranking or score for them.

This is, however, precisely one of the areas that European AI regulation expressly addresses. Point 4 of Annex III to the AI Act lists among high-risk systems systems intended for the recruitment or selection of natural persons, including systems used to analyze and filter job applications and to evaluate candidates.

A high-risk system therefore need not be only one that itself decides whom the employer hires. Automated CV screening or candidate evaluation alone may fall within the regulated area. At the same time, the specific purpose and manner of use of the system must be assessed, because Article 6(3) of the AI Act provides, under certain conditions, exemptions for some systems that perform narrowly defined supporting or procedural tasks.

One of the most important questions when using AI in HR is: who is legally liable for an algorithm's error?

AI does not bear legal liability. If an employer uses AI when selecting a candidate, evaluating an employee or deciding on an employment relationship, it cannot escape its responsibility by pointing out that “the system decided”.

Recent case law outside employment law illustrates this well. In its judgment Ref. No. 10 As 81/2026, the Czech Supreme Administrative Court highlighted a situation in which a procedural submission relied on a number of purported case-law references generated by AI that were incorrect or did not exist. In this context, the court stressed that an attorney, as a legal professional, is responsible for their submission and also for any “hallucinating” AI argumentation if AI was used in preparing it.

In other words: AI may recommend a decision, but responsibility for the decision remains with a human, or rather with the employer.

For high-risk systems, Article 14 of the AI Act sets out requirements for human oversight. The person performing oversight must be able to recognize the system's limitations, correctly interpret its output and, where necessary, disregard or reverse the result or intervene in the system. It is therefore not enough to create a process in which AI evaluates candidates and an HR officer then simply clicks an “approve” button.

A problem may arise, for example, with a system that automatically ranks candidates by a score from 1 to 100, with the HR officer inviting only those with the highest scores to an interview. If the HR officer does not actually review the algorithm's output but merely adopts it mechanically, it is questionable whether such involvement can be considered genuine human decision-making.

This issue is also significant from a GDPR perspective. Article 22 of the GDPR gives individuals the right not to be subject to a decision based solely on automated processing of personal data, including profiling, which produces legal effects concerning them or similarly significantly affects them.

The importance of this provision was further underlined by the Court of Justice of the EU in its judgment in Case C-634/21 SCHUFA Holding. The court held that the automated generation of a score may, in certain circumstances, itself constitute automated individual decision-making where a third party draws strongly on that score in reaching its decision.

The practical conclusion for HR is this: the formal presence of a human at the end of the decision-making process may not be sufficient. What matters is whether the human genuinely assesses the AI output, or merely confirms a result that was in fact produced for them by the algorithm.

The most sensitive situation arises when AI begins to be used as a basis for terminating an employment relationship.

Such use of AI falls within the area of high-risk systems under point 4 of Annex III to the AI Act. Alongside European regulation, however, the Czech Labor Code must also be applied, because an algorithm's decision in itself is not a ground for dismissal. If an employer terminates employment by notice under Section 52 of the Labor Code, the conditions of a specific statutory ground for dismissal must be met.

AI can therefore serve as a tool for gathering and processing supporting material. It cannot, however, replace the employer's legal responsibility for the decision itself to terminate employment.

European rules on platform work offer an interesting view of future developments. Directive (EU) 2024/2831 regulates, among other things, so-called algorithmic management, i.e. the use of automated monitoring and decision-making systems to organize work, evaluate performance or take decisions that significantly affect persons performing platform work.

Of particular importance is the rule that certain decisions concerning the restriction, suspension or termination of the contractual relationship should not be left to an algorithm alone. Workers should be able to obtain an explanation of the decision and request its human review.

Gamification of the workplace can also be a specific example. An employer may use point systems, employee leaderboards or rewards for meeting certain metrics as a motivational tool.

The use of game elements in itself need not be a problem. The risk arises, however, when the game score is linked to AI and begins to be used for significant employment decisions. AI may, for example, continuously track an employee's activity, speed of task completion, number of requests handled or engagement in internal systems, and convert all of this data into a single performance score.

A tool meant to support motivation can thus become a tool for intensive management and monitoring of employees. From a legal perspective, the decisive factor is therefore not only whether the system contains a “game” element. What matters is what data it collects, how it profiles employees and what decisions its results are used for.