The EU AI Act and Works Councils: Consultation Obligations for Workplace AI
As of 2 August 2026, the EU AI Act's obligations for high-risk AI systems apply — and the list of high-risk systems is, to a striking degree, a list of HR technology. CV screening, candidate ranking, task allocation, performance evaluation, monitoring: if your organisation deploys AI anywhere in the employment lifecycle, you are now operating regulated technology. And before that technology reaches the workplace, workers and their representatives must be told. For multinationals with works councils, the AI Act has quietly become a consultation topic — arguably the fastest-growing one in European labour relations.
This article explains what the AI Act actually requires of employers as deployers of AI, how those requirements interact with existing works council rights in Germany, France, and the Netherlands, and how to sequence consultation so that AI rollouts are not derailed by an obligation nobody mapped.
What the AI Act says about AI at work
Regulation (EU) 2024/1689 — the AI Act — entered into force on 1 August 2024 and has been applying in stages. The prohibitions took effect on 2 February 2025, obligations for general-purpose AI models followed on 2 August 2025, and the core compliance regime for high-risk systems listed in Annex III has applied since 2 August 2026. Three parts of the Act matter directly for employment.
Some workplace AI is banned outright. Since February 2025, AI systems that infer emotions in the workplace are prohibited, save for narrow medical and safety exceptions. So is social scoring and the untargeted scraping of facial images. If any tool in your HR or workplace-monitoring stack claims to read employee sentiment, stress, or engagement from video, voice, or biometric data, it needs urgent legal review — the penalty ceiling for prohibited practices is the highest in the Act, up to €35 million or 7% of worldwide annual turnover.
Most employment AI is "high-risk". Annex III designates as high-risk the AI systems used for recruitment and selection — placing targeted job ads, filtering applications, evaluating candidates — and those used to make decisions on promotion and termination, to allocate tasks based on individual behaviour or traits, and to monitor and evaluate performance. High-risk classification triggers duties for the provider (the vendor) and, crucially, for the deployer — the employer using the system.
Deployers have their own obligations. Under Article 26, an employer deploying high-risk AI must use the system in accordance with its instructions, assign competent human oversight, ensure input data is relevant and sufficiently representative, monitor operation, and retain logs. And Article 26(7) contains the sentence that brings labour relations into the picture: before putting a high-risk AI system into service at the workplace, employers shall inform workers' representatives and the affected workers that they will be subject to it. Not after go-live. Before.
Why "inform" becomes "consult" in practice
Article 26(7) is drafted as an information duty. But the AI Act explicitly preserves more protective national law: it does not displace member-state labour law or collective agreements, and it invites them to go further. In the countries where most European works councils sit, national law already goes considerably further — which means that for a works-council country, the AI Act's information duty is the floor, and genuine consultation or even co-determination is the operating reality.
Germany. The Betriebsverfassungsgesetz was amended in 2021 specifically to address AI. The works council must be informed and consulted on the planning of work processes involving AI (§90 BetrVG), and it may call on an external expert in AI matters without the usual necessity test (§80(3)). Where AI is used to set selection guidelines for hiring, transfers, or dismissals, co-determination applies (§95). And §87(1)(6) — co-determination over technical systems capable of monitoring behaviour or performance — catches most AI deployments in practice, because German case law reads "capable of monitoring" broadly. The practical consequence: workplace AI in Germany normally requires a works agreement (Betriebsvereinbarung) negotiated with the works council before rollout. Our Germany country landscape covers the co-determination framework in more detail.
France. The CSE must be informed and consulted before the introduction of new technologies and on any significant change to working conditions — a category that comfortably includes algorithmic management and AI-assisted decision tools. French case law in the last two years has underlined that deploying such tools without completing CSE consultation exposes the employer to suspension of the project. See the France country landscape for the CSE's consultation architecture.
Netherlands. The Dutch works council (ondernemingsraad) holds an advice right over major technology investments and, more pointedly, a consent right (instemmingsrecht) under Article 27 WOR for systems that process employee personal data or monitor employees. An AI system that evaluates performance almost always lands in consent territory: without OR consent or substitute court approval, the decision is voidable. The Netherlands country landscape sets out how the advice and consent tracks differ.
Layer on top of this the GDPR — Article 22 limits on solely automated decisions with significant effects, and Article 88 workplace-specific protections — and the direction of travel in the Platform Work Directive (EU) 2024/2831, which builds a full algorithmic-management regime (human review of automated decisions, transparency towards workers' representatives) that member states must transpose by 2 December 2026. The platform directive formally covers platform work only, but several member states have signalled an appetite to extend algorithmic-management rules more widely. Employers who treat the AI Act as the whole story will keep being surprised.
What does this mean for a multinational rollout?
The uncomfortable feature of workplace AI, from a programme-management perspective, is that it is usually deployed centrally — one global tool, one vendor contract, one go-live plan — while the consultation obligations it triggers are national and local. A recruitment-screening tool switched on for all of Europe simultaneously may require: information to workers and representatives everywhere (AI Act), a negotiated works agreement in Germany, completed CSE consultation in France, OR consent in the Netherlands, and — where the deployment forms part of a broader transnational change affecting workers in two or more member states — information and consultation of the European Works Council as well.
That EWC layer deserves emphasis. The revised EWC directive broadens what counts as a transnational matter and, from 2029, requires management to respond to EWC opinions with written reasons before implementing the decision. A pan-European AI deployment is close to a textbook transnational matter. Multinationals that have an EWC should expect it to ask about AI; multinationals that plan AI rollouts as purely local IT projects should expect the question to arrive mid-deployment, when it is most expensive.
A practical sequencing plan
Teams that manage this well treat workplace AI like any other consultation-bearing change programme, with four disciplines.
1. Inventory and classify. Build a register of every AI system that touches the employment relationship — including features embedded inside HR suites you already run, which vendors have been quietly upgrading into scope. Classify each against the Act: prohibited, high-risk, limited-risk, minimal. This register is also the natural artefact to share with representatives, and being able to produce it changes the tone of every subsequent conversation.
2. Map the obligation per country, per system. Inform, consult, or obtain consent — the answer differs by jurisdiction and by what the system does. The mapping determines your critical path: a Dutch consent right or a German works agreement can take months, and neither can be compressed by announcing a go-live date first.
3. Sequence consultation before deployment, and evidence it. Article 26(7) says before. So do the national regimes. That means representative engagement belongs in the project plan alongside vendor selection, not as a communications task at the end. And because AI deployments are precisely the kind of decision that gets challenged, the record matters: what was shared with which body, when, what questions were raised, what the employer answered, what was agreed. If a works council or court asks you to evidence the process a year later, that record should be retrievable, not reconstructable.
4. Prepare for the ongoing relationship. AI systems change — models are retrained, features added, thresholds moved. German works agreements on AI typically include review clauses; Dutch consent covers the system as described. Material changes reopen the conversation. Treating AI governance as a standing agenda item with representative bodies, rather than a one-off project, is both legally safer and better relations.
The window to get ahead of this
Enforcement of the high-risk regime is only beginning, and most works councils are still building their own literacy on AI — many, in Germany, by exercising their statutory right to bring in experts. Employers currently have a window in which proactive transparency is cheap and buys durable goodwill. That window closes the first time a representative body discovers a deployed system it was never told about: from that point, every subsequent technology conversation starts from mistrust, and the formal machinery — information requests, expert appointments, injunction threats — gets used instead.
The organisations handling this best have stopped treating AI consultation as an exotic new obligation. It is ordinary works council management applied to a new subject matter: know your bodies and their rights, prepare honest documentation, consult before deciding, record everything. The subject is new; the discipline is not.
Graylark LRM gives multinational employers the operational infrastructure for exactly this: representative and body tracking across countries, structured consultation workflows with deadlines and ownership, meeting and document records linked to each process, and an audit trail produced from the system of record — so an AI rollout's consultation critical path is visible from day one.
See how Graylark handles works council management