Annex III point 4 lists AI systems intended to be used to allocate tasks based on individual behaviour or personal traits or characteristics, or to monitor and evaluate performance and behaviour of persons in work-related relationships, as high-risk use cases, and it also lists systems intended to be used to make decisions on promotion, termination, task allocation and self-employment access on the basis of personal traits, behaviour or characteristics. Those functions must be tested against the listing even where the system is described as a productivity or scheduling tool. Article 5 screening for emotion inference and biometric categorisation, worker information and consultation under employment law, and data protection duties must be tested separately. This guide does not give employment law advice and does not decide any specific workplace deployment.
Which worker management functions are listed?
Annex III point 4 lists, in outline, systems intended to allocate tasks based on individual behaviour or personal traits or characteristics, systems intended to monitor and evaluate the performance and behaviour of persons in work-related contractual relationships, and systems intended to make decisions on promotion, termination, task allocation and access to self-employment on the basis of personal traits, behaviour or characteristics. A scheduling system that assigns work by availability alone is assessed differently from a system that assigns work by inferred behaviour or traits, and a system that records output for payroll is assessed differently from a system that evaluates behaviour and influences promotion or termination. The record should therefore describe the function in operational terms: what data are used, what inference is made, what allocation or evaluation follows, and what consequence it has for the worker. A productivity dashboard, a monitoring tool and a decision tool can be the same product with different functions, and each function needs its own assessment.
How does monitoring differ from task allocation?
Monitoring and task allocation are listed as distinct functions, and the distinction matters for the assessment. Monitoring observes and evaluates what a worker does, how they do it and how they behave, and it can affect privacy, dignity and the employment relationship even where no formal decision follows. Task allocation decides what work a person does, when and under what conditions, and it can affect earnings, hours, safety and career progression. A system that does both should be assessed for both functions, and the oversight, information and consultation arrangements should address each. The deployer should also consider whether monitoring is continuous or episodic, whether it is visible to the worker, and whether the worker can understand and challenge the data and inferences used. Covert or continuous monitoring that the worker cannot understand or challenge raises fundamental rights and employment law questions that go beyond the AI Act classification, and those questions need qualified advice before deployment.
For the broader regulatory framework, see healthcare and biometrics under the AI Act.
What must be screened under Article 5?
Workplace systems must be screened under Article 5 before classification, with particular attention to emotion inference and biometric categorisation. A system that infers a worker's emotions from facial expressions, voice or other biometric data in the workplace is within the Article 5 prohibition on emotion inference, except for medical or safety reasons, and the exception must be tested on the facts. A system that categorises workers by biometric data to infer sensitive attributes can engage the biometric categorisation prohibition. A system that exploits vulnerability, manipulates behaviour or carries out social scoring can engage other prohibitions. The screen should examine every data source and inference, including keystroke, location, communication and video data, and should record the conclusion for each prohibition. A workplace tool that is lawful in one function can be prohibited in another, and the prohibited function must be removed or redesigned, not carried forward into conformity work.
What information and consultation duties apply?
Deployers of high-risk systems must inform workers that they are subject to the use of a high-risk system, and must inform workers' representatives where employment law requires consultation, before the system is used. Data protection law requires information about the processing, its purposes and the worker's rights, and employment law in the Member State concerned can require consultation with workers' representatives or a works council before a monitoring or evaluation system is introduced. The deployer should therefore plan a single information and consultation process that covers the AI Act, data protection and employment law together, with a clear timetable, a named owner and a record of what was communicated and what was agreed or objected to. Workers should be told what data are collected, what inferences are made, what decisions follow, how they can challenge a decision and how they can raise a concern. A deployment that begins before consultation is complete risks an employment law challenge even where the AI Act file is in order.
What oversight should a workplace deployment have?
Oversight in the workplace means a competent person, with time and authority, reviews the system's output before a consequential decision is taken, considers information beyond the output, and can decide differently from the system. The overseer should understand the system's limitations, its error rates and its potential for bias, and should record the decision, the information considered and any override. The deployer should monitor the system's operation, including whether it treats workers consistently, whether its outputs are followed or overridden, and whether it produces unintended consequences for safety, workload or wellbeing. Workers should have a route to raise concerns about the system, and those concerns should be recorded and addressed, with feedback to the worker concerned. An oversight arrangement in which a manager simply confirms the system's output, without time or information to assess it, is not effective oversight, and the record should be able to show that oversight was real.
Continue with employment recruitment and selection.
What does this guide not decide?
This guide does not decide whether any workplace system is high-risk or prohibited, does not determine whether a monitoring or allocation practice is lawful under employment or data protection law, and does not give advice on consultation, collective bargaining or individual employment rights. Those questions depend on the system's function, the data and inferences, the employment relationship and the law of the Member State concerned, applied to the specific facts with qualified advice. The guide also does not assess the validity of any performance or behaviour measure. Where a worker management system is to be deployed, confirm the classification, prohibition, consultation, data protection and oversight position with a competent authority or legal adviser before the system affects workers.
Frequently asked questions
Is a scheduling tool high-risk?
It depends on its function. A tool that allocates tasks based on behaviour or personal traits, or that monitors and evaluates performance, is listed in Annex III point 4. A tool that schedules by availability alone, without those inferences, is assessed differently.
Must workers be consulted before a monitoring system is introduced?
Employment law in the Member State concerned can require information and consultation with workers or their representatives, alongside AI Act and data protection information duties. The consultation process should be planned and recorded before deployment.
Can emotion inference be used to monitor workers?
As a general rule, no. Article 5 prohibits emotion inference in the workplace except for medical or safety reasons, and the exception must be tested on the facts with qualified advice.
Continue through this AI Act cluster
This article is an informational overview, not a legal determination. Check the current legal text and official implementation guidance for your system and role.
Consolidated Regulation (EU) 2024/1689, as amended ↗AI Act Service Desk implementation timeline ↗