Artificial Intelligence in Disciplinary Proceedings: The Limits of Automated Decision-Making in Labor Law

Can an algorithm flag employee misconduct – or even influence a disciplinary sanction? As AI becomes increasingly embedded in workplace monitoring and employee assessment, the distinction between technological assistance and human decision-making is becoming a critical employment law issue.

The development of artificial intelligence (“AI”) is bringing about significant changes in labor relations, including with regard to the monitoring and evaluation of employees.

However, the use of AI systems to identify potential disciplinary violations raises the issue of whether these technologies are compatible with the provisions of the Labor Code and with European legislation on data protection and artificial intelligence.

According to Article 247 of the Labor Code, the employer has disciplinary authority, and a disciplinary offense involves the willful commission of a work-related act that violates the law, internal regulations, an individual or collective employment contract, or the lawful instructions of superiors.
Article 250 of the Labor Code requires the employer to determine the sanction based on the severity of the offense and mandates that the employer take into account the circumstances under which it was committed, the degree of culpability, the consequences of the offense, the employee’s overall conduct, and any prior disciplinary sanctions.

This latter provision is important in relation to AI because an algorithm may, for example, automatically classify a behavior as “high risk” or assign an employee a certain performance score; however, such a classification is not equivalent to a legal analysis of culpability.

Therefore, an AI system can identify suspicious behavior or an apparent violation of a rule, but it cannot replace the employer’s legal analysis regarding the existence of the misconduct and culpability.
AI can assist in the analysis, but the decision must allow for an assessment of the specific circumstances of each case.

An important reference point is Article 22 of Regulation (EU) 2016/679, which governs decisions based solely on automated processing.

In this regard, in Case C-634/21 SCHUFA Holding, the CJEU held that an automatically generated score may fall within the scope of an automated decision when it substantially determines the subsequent decision made by another person or entity. This principle is also relevant in the context of employment relationships, as formal intervention is not necessarily sufficient if it is limited to confirming the algorithm’s result without the employer conducting its own analysis.

Thus, the Disciplinary Investigation Committee may use AI when drafting the proposal for disciplinary action to be submitted to the employer, however, the algorithmic result should be purely advisory in nature, and committee members are obligated to review, understand, and directly endorse the final proposal.

The employer, through its legal representative, may use AI in the process of justifying the disciplinary sanction; however, the decision must be based on a direct and rigorous analysis conducted by the employer, with the algorithmic result playing only a subsidiary role in verifying or confirming the employer’s own assessment.

In conclusion, artificial intelligence can assist in the disciplinary process but cannot assume the authority to impose disciplinary sanctions. Responsibility for the legality of the sanction, compliance with procedure, and the individualization of the sanction remains with the employer, and the employee must be guaranteed the right to a defense and the opportunity to challenge the automatically generated data or conclusions.