Annex III employment and workers management: recruitment, task allocation and self-employment

Which recruitment uses does point 4 cover?

Point 4 covers recruitment and selection uses, including placing targeted job advertisements, screening or filtering applications, and evaluating candidates during interviews or tests. The coverage starts before a human interviewer sees a candidate: an advertising system that targets vacancies, a screening tool that ranks applications, and a test evaluation tool that scores a candidate are all within the point's description where they are intended for those purposes. A general job board that only lists vacancies, without screening, filtering or evaluation, follows a different analysis. The record should describe the recruitment stage, the decision the system influences, and the criteria it applies, because a tool used at the advertising stage and a tool used at the final selection stage raise different evidence questions.

Which workers management uses does point 4 cover?

Point 4 also covers systems used to make decisions affecting terms of work-related relationships, including promotion and termination, to allocate tasks based on individual behaviour or personal traits or characteristics, and to monitor and evaluate performance and behaviour of persons in work-related contractual relationships. Task allocation based on behaviour or traits, productivity monitoring, and performance scoring are the core illustrations. A system that allocates tasks by availability or skill, without using behaviour or personal traits, and a system that aggregates team-level statistics without evaluating individuals, need careful wording analysis, because the point turns on the basis of allocation and on individual evaluation. The record should state the basis, the individual affected, and the consequence of the decision.

For the broader regulatory framework, see high-risk AI systems.

What about self-employment and platform work?

Point 4 covers systems intended to be used to determine access to self-employment, to revoke access, or to allocate tasks to self-employed persons. Platform allocation systems that assign work, set access, or evaluate self-employed workers are therefore within the point's description where they perform those functions. The employment status label used by the platform does not decide the analysis. What matters is whether the system determines access or allocates tasks to a person working under a self-employment arrangement. A system that only lists available work, without determining access or allocating tasks, follows a different analysis, but the boundary should be evidenced from the actual allocation logic, not from the platform's description of itself.

How do employment law and worker consultation interact?

Employment law, works council and worker consultation rules, equality law, and data protection law apply alongside the AI Act to employment systems. Deployers must inform workers and their representatives about the use of a high-risk system under Article 26(7), where that paragraph applies, and must consider consultation duties under national employment law. The AI Act classification does not replace those duties, and a lawful AI Act classification does not make a discriminatory or unlawful employment practice lawful. The record should cross-reference the employment law advice, the consultation carried out, and the data protection assessment, so that the AI Act conclusion is not mistaken for a complete employment law clearance.

What evidence matters for an employment classification?

An employment classification record should describe the decision affected, the criteria and data used, the validation of the method, the error and bias testing across relevant groups, the human review design, and the route for a worker or candidate to challenge the outcome. Bias testing is particularly important, because recruitment and performance systems can reproduce historical patterns in training data. The record should state what testing was done, on what data, with what results, and what limits remain. Where the Article 6(3) filter is relied on, the documentation and registration duties in Article 6(4) and Article 49(2) apply, and the profiling override must be addressed directly, because many employment systems perform profiling.

Continue with Article 6 classification routes.

What does this guide leave open?

This guide explains point 4 in outline. It does not decide whether any particular recruitment, allocation, monitoring or platform system is high-risk, whether the Article 6(3) filter applies, or whether an employment practice is lawful. Those questions need the decision affected, the criteria, the data, and the human review design, read against the consolidated Annex III wording and the applicable employment and equality law. Employment systems affect livelihoods, so borderline cases should be confirmed with a competent authority or legal adviser, and worker representatives should be involved through the routes the applicable law provides.

Frequently asked questions

Continue through this AI Act cluster

• high-risk AI systems
• Article 6 classification routes
• Article 6(3) filter and its limits
• deployer responsibilities
• AI Act in finance, insurance and infrastructure
• provider responsibilities