An AI project rarely stalls on technology. It stalls in month four, when it turns out the system processes employee data and the works council was never consulted. By then the build is finished, the rollout is scheduled, and the decision is legally challengeable.
Works council consent for AI is the power under Article 27 of the Dutch Works Councils Act (WOR) to block a proposed decision when an AI system processes employees' personal data or is capable of observing their attendance, behaviour or performance. Without consent such a system may not be put into use, however far the implementation has progressed.
This guide covers which AI use cases fall under which article, what happens if you skip the council, and how to run the project so consent doesn't cost you time. This is general information, not legal advice.
When does the works council have a right of consent?
Dutch law requires organizations with fifty or more employees to establish a works council. That is precisely the group where most AI implementations happen, and two parts of Article 27 are almost always relevant.
| Basis | What it covers | Typical AI use case |
|---|---|---|
| Article 27(1)(k) | rules on processing and protecting employees' personal data | AI processing HR files, applications or employee email |
| Article 27(1)(l) | facilities aimed at or suitable for observing or monitoring attendance, behaviour or performance | AI measuring productivity, analysing calls or scoring work |
| Article 25(1) | major organizational or technological changes | company-wide AI rollout that changes roles or processes |
Pay attention to the words "suitable for" in point (l). No intent to monitor is required. A system that summarises customer service call recordings is technically suitable for assessing employee performance, even when that isn't your purpose. That suitability alone triggers the right of consent.
This makes the scope wider than most project leads assume. An AI assistant on the internal knowledge base that logs who searches for what, a planning system that tracks individual cycle times, or a tool that scores tickets: all three touch these grounds.
What is the difference between advice and consent?
The two get used interchangeably, but the difference decides how much room you have.
Under the right to advise (Article 25) you must ask the council for advice before deciding and weigh that advice seriously. If you depart from it, you have to justify the departure and wait a month before implementing, so the council can go to the Enterprise Chamber if it wants. You do get to decide in the end.
Under the right of consent (Article 27) it works differently: without consent there is no decision you may implement. If the council refuses, you can ask the subdistrict court for substitute authorisation, but implementation is on hold until that ruling.
For AI projects this means the question "which right applies" belongs at the start of the project, not the end. If the system processes employee personal data or can assess their work, assume consent is required.
[ TIME SAVED ]
Save 20 hours per week on redoing an AI implementation that was rolled out without works council consent
What happens if you skip the works council?
The sanction is more direct than under most other regulation. A decision that required consent but didn't get it can be declared void by the council. The decision then has no legal effect: you may not use the system, and the council can enforce compliance through the courts.
In practice that creates three problems, all of which cost money. Your rollout sits still while licences keep running. Your negotiating position with the council has weakened, because you're repairing something rather than proposing it. And the loss of trust carries into the next project, where you need the same council again.
On top of that, processing employee data also falls under the GDPR, with its own requirements on legal basis, proportionality and transparency. The works council route and the privacy route run in parallel; neither replaces the other.
What does the AI Act change?
The AI Act adds a separate information duty. Employers deploying a high-risk AI system in the workplace must inform workers' representatives and the affected employees in advance that they will be subject to the use of that system.
Two caveats matter. First, this is an information duty and not a right of consent: the AI Act doesn't replace the Dutch Works Councils Act, it sits alongside it. Second, the obligations for deployers of Annex III high-risk systems moved to 2 December 2027 under the Digital Omnibus.
So the delay applies only to the European layer. The WOR obligations apply now, with no transition period, and for most AI implementations in the Netherlands they are the binding framework. Waiting until December 2027 means waiting on the wrong law. For the full regulatory picture, see our pillar on AI legislation in the Netherlands and the EU AI Act.
The AI Act also sets requirements for the knowledge of staff working with AI systems. Our explainer on the AI literacy requirement covers what that involves.
How do you run an AI project with the works council?
Involve the council early and you rarely lose time. Involve it late and you always do. Five steps work in practice.
- Determine in the design phase which right applies. Does the system process employee data, or can it assess work? Record the conclusion in writing, with the reasoning.
- Brief the council informally before a formal request lands. A council that understands the problem before it has to judge a solution asks better questions and decides faster.
- Bring a concrete proposal, not an intention. Which data, which purpose, what retention period, who has access, what is explicitly not measured, and how the council can verify that.
- Fix the limits in the rules themselves, not in an explanatory note. For example: no individual performance scores, no use in appraisals, no link to the HR system.
- Agree an evaluation moment. Consent for an indefinite period is harder to obtain than consent with a review after six months.
Point four is where most projects gain. Resistance is rarely aimed at AI itself, but at the fear that data will later be used for something else. Fixing that boundary firmly takes the heat out of the discussion. The same logic applies to uncontrolled use of AI tools by employees; our guide on a shadow AI policy covers that side.
Which use cases usually need consent?
A rough split helps you assess your own portfolio.
Almost always requires consent: AI in recruitment and selection, performance monitoring, analysis of employee customer calls, individual-level scheduling, and any system processing employee email or chat. Recruitment carries additional rules too; see our explainer on AI in recruitment.
Usually does not: AI processing only customer data with no traceability to employees, document classification without user logging, demand forecasting, and generic text generation without storing who entered what.
Borderline: an AI assistant on the internal knowledge base that logs search behaviour, and automation that measures cycle time per step. Here it depends on the setup: what you don't log can't be used for assessment. A technical choice is often cheaper than a legal discussion. How to bring staff along in the use itself is covered in our guide on training employees on AI tools.
Conclusion: settle the works council question in design
The right of consent isn't an obstacle; it's a design question. Decide in week one which data the system needs and which it explicitly won't collect, and you put forward a proposal the council can assess rather than block.
Skip that question and you'll meet it at delivery, paying in delay and in trust. In our AI consulting work we build the employee-data question into the design by default, because a system that records less about staff often works just as well and is far easier to introduce.