Does the works council need to approve AI tools in the workplace?

The HR director of a Dutch logistics company is about to roll out two AI tools: a system that pre-screens incoming job applications and a dashboard that visualises planner productivity. The licences are nearly signed. Then the chair of the works council calls: "Doesn't this require our consent?" It is the right question, and the answer decides whether the project proceeds or goes back to the drawing board.

The direct answer: in many cases, yes, the works council must consent to AI tools in the workplace. The legal basis is Article 27(1) of the Dutch Works Councils Act (Wet op de ondernemingsraden, WOR). As soon as an AI tool processes personal data of employees (sub k) or is aimed at or capable of observing or monitoring the presence, behaviour or performance of employees (sub l, the classic employee monitoring provision), the decision to adopt, amend or withdraw such an arrangement requires the works council's prior consent. If the employer proceeds without it, the works council can invoke the nullity of the decision. This article sets out when the consent right applies, what the EU AI Act adds on top, and how HR and works councils can organise this well in practice.

What Article 27 of the Works Councils Act covers

Article 27(1) WOR lists the subjects for which the employer needs the works council's prior consent. For AI in the workplace, two items are almost always relevant:

• Sub k: arrangements concerning the processing and protection of personal data of the people working in the organisation. Practically any AI tool that touches employee data falls within this scope, from a CV screening tool to an HR chatbot that logs conversations.
• Sub l: arrangements concerning facilities that are aimed at or capable of observing or monitoring the presence, behaviour or performance of employees. This is the employee monitoring provision, and its wording is deliberately broad.

Two legal nuances matter. First, the consent right applies to arrangements: decisions of general application that affect a group of employees. Buying software is not in itself an arrangement, but introducing an AI tool for recruitment, appraisal or scheduling almost always is, because it comes with policy choices about who uses the system, which data go in and what happens with the outputs. Second, sub l says "aimed at or capable of". The employer's intention is not decisive. An AI planning tool that, as a by-product, records how quickly each employee completes tasks is capable of performance monitoring and therefore subject to consent, even if nobody plans to use the data that way.

Alongside the consent right, the advice right can also apply: Article 25(1)(k) WOR gives the works council a right to advise on the introduction or significant change of an important technological facility. In an organisation-wide AI rollout, the advice and consent tracks often run in parallel.

When does an AI tool require works council consent?

A practical three-question test:

1. Does the tool process personal data of employees or applicants? Think of CVs, appraisals, chat logs, login times, location data or productivity metrics. If yes, sub k is in play.
2. Can the tool be used to track presence, behaviour or performance? Even if that is not its purpose. If yes, sub l is in play.
3. Is there an arrangement of general application? A pilot with three volunteers may still fall outside, but once the tool becomes part of the HR or work process for a group of employees, there is an arrangement.

In practice this means that CV screening and applicant ranking, AI-assisted appraisal and promotion systems, productivity and output dashboards, AI monitoring of email or customer calls, and algorithmic shift scheduling almost always require consent. A generic AI assistant that employees use to draft texts sits in a greyer zone: not a monitoring system as such, but as soon as prompts and usage are logged per employee and traceable, sub k enters the picture after all.

What if the employer bypasses the works council?

If the employer implements a decision that required consent without obtaining it, the works council can invoke the nullity of that decision in writing. It must do so within one month after the employer communicated the decision or after the council noticed it was being implemented (Article 27(5) WOR). The decision is then treated as never taken, and the works council can ask the subdistrict court to enforce compliance. Conversely, if the works council withholds consent, the employer can ask the subdistrict court for substitute permission (Article 27(4) WOR). The court then weighs whether the refusal is unreasonable or whether compelling business interests require the decision. The lesson for HR is simple: skipping the consent procedure does not save time, it produces a project that stalls for months.

The Dutch Data Protection Authority (Autoriteit Persoonsgegevens, AP) watches this space too. Employee monitoring engages the GDPR, often including the obligation to carry out a data protection impact assessment (DPIA) in advance. The AP is also the coordinating supervisor for algorithms and AI in the Netherlands, alongside the RDI for the more technical side. In monitoring cases, a breach of the Works Councils Act and a GDPR violation regularly go hand in hand, with the GDPR carrying the possibility of a fine imposed by the supervisor.

What the EU AI Act adds

The Works Councils Act governs employee participation; the EU AI Act governs the quality and transparency requirements for the system itself. Three points matter for HR applications.

High-risk obligations from 2 December 2027. AI systems used for recruitment and selection, decisions on promotion and termination, task allocation, and monitoring and evaluating employees are listed in Annex III of the AI Act and qualify as high-risk. Regulation (EU) 2026/1744 fixes 2 December 2027 as the date for the core obligations concerning this category. For employers deploying such systems, one obligation connects directly to employee participation: Article 26(7) AI Act requires the employer to inform workers and their representatives before putting a high-risk AI system into use in the workplace. A works council that is already at the table today will formally be the counterpart then.

Transparency since 2 August 2026. For an HR chatbot intended to interact directly with employees or applicants, Article 50(1) places the disclosure-by-design duty on the provider, unless the AI nature is already obvious in context. An employer deploying a procured chatbot should verify that feature and use it according to the provider's instructions. The employer may also have separate deployer duties under Article 50(3) or (4), depending on the use case.

AI literacy applies today. Article 4 AI Act has applied since 2 February 2025 and expects organisations to ensure that staff working with AI have a sufficient level of AI literacy. It is not a provision with its own fine, but it is the foundation for meaningful human oversight (Article 14) and exactly the kind of commitment a works council can secure in a consent procedure: who gets trained before the tool goes live? Platforms such as LearnWize are built for that purpose.

The practical route: involve the works council early and share your AI register

Organisations where this runs smoothly do three things differently. First, they involve the works council before the vendor is selected, not after. A consent request about a fait accompli feels like a formality to any works council and invites refusal; a council that helped shape the selection criteria rarely blocks the outcome.

Second, they share their AI register with the works council. An up-to-date overview of all AI systems in the organisation, including purpose, data flows and risk class, lets you show the council the full picture at once and assess per system whether Article 27 applies. How to build such a register is covered in our article on the AI inventory under the AI Act. The register thus doubles as the natural agenda item for the periodic general consultation between employer and council.

Third, they record the agreements as a proper arrangement: purpose limitation, which data the system uses, who sees the outputs, how employees can object, an evaluation moment after six or twelve months, and the commitment that the tool will not be extended into monitoring without renewed consent. Organisations that tie this into a broader governance structure, for instance along the lines of ISO 42001, will find most of the works council file already in place. Organisations that want support in setting up this process, from AI register to consent request, can turn to Embed AI.

The core message for HR directors: treat the consent right not as an obstacle but as a built-in quality check. And for works council members: you do not have to wait until the AI Act's high-risk obligations take effect. Article 27 of the Works Councils Act gives you a firm seat at the table today.