Annex III names employment and worker management directly, including systems used to evaluate candidates.
Why it matters when the plan changes
Annex III is the list that decides whether a system carries the heavier set of obligations, and employment is on it explicitly, naming systems that place targeted job advertisements, filter applications and evaluate candidates. That makes classification a question to answer deliberately rather than a matter of opinion, and it makes documentation and oversight design questions to build toward now, ahead of the 2 December 2027 date when the obligations start applying, rather than things to assemble if someone asks.
The tension is between building for the requirements and claiming to have met them. The two are different, and the distance between them is an assessment. A supplier that has built documentation, traceability and human oversight ahead of that date has done real work, and it has not thereby been found compliant; compliance is claimed only after assessment, and saying otherwise is a claim that cannot be evidenced.
In practice
A procurement questionnaire asks whether a supplier is AI Act compliant. The accurate answer names what has been built, what it is being built against, which obligations would apply if the system is classified under Annex III, and states that compliance is claimed only after assessment. That answer survives the next question; a simple yes does not.
Evidence
Annex III lists employment and worker management among the high-risk use cases, including systems that evaluate candidates.
Annex III, European Union Artificial Intelligence Act (2024)The obligations for systems classified as high risk under Annex III start applying from 2 December 2027.
Implementation timeline, European Union Artificial Intelligence Act (2024)
What it cannot tell you
Annex III determines which obligations apply; it does not itself specify how to meet them, and it does not certify that any particular system does. Classification is a starting point for assessment, not a verdict on a system's quality or fairness, and the Annex is silent on the substance of accuracy, bias testing or oversight design.
Questions
The use cases classified as high risk under the Act, grouped by area. Annex III, European Union Artificial Intelligence Act (2024), names employment, worker management and access to self-employment as one group, covering systems intended to place targeted job advertisements, filter applications and evaluate candidates.
Risk management, data governance, technical documentation, record keeping, transparency toward users, human oversight, and accuracy and security requirements, together with the conformity work needed to demonstrate them. The detail sits in the Act itself rather than in the Annex, which only establishes the classification.
The official implementation timeline gives 2 December 2027 for systems classified as high risk under Article 6(2) and Annex III. Dates in this area have been adjusted before, so the current official timeline is the source to check rather than a figure remembered from a briefing.
No such claim is made. Documentation, traceability and human oversight are being built ahead of the 2 December 2027 Annex III date, and compliance is a finding that follows an assessment rather than a statement a supplier makes about itself. The distinction is stated plainly.
No. It classifies certain uses as high risk and attaches obligations to them, chiefly around documentation, transparency and meaningful human oversight. The effect is to require that such systems be built and governed in a particular way, not to prohibit the category.