Works Council and AI: What Requires Co-Determination and Why Timing Matters

The AI tool has been selected, the licenses are paid for, and the rollout is on the calendar. On Wednesday, the email goes to the works council: "Please approve by Friday." For the works council, it is the same week from the other direction: it is supposed to approve a system it is seeing for the first time. The answer is predictable – not yet.

So is the works council a brake on AI? Current data show no measurable braking effect: the Institute for Employment Research (IAB) found no statistically significant relationship between the existence of a works council and the use of generative AI [1]. A survey by the Weizenbaum Institute points to where things get stuck: 40 percent of the works councils surveyed are always or often involved when new technologies are planned, while the majority are involved only sometimes or not at all [2]. Often, the works council simply comes in too late.

This article explains what rights a works council in Germany has with regard to AI, what the EU AI Act changes for AI in the workplace from December 2027, and how employers and works councils can reach a solution in four steps that both can support. It is written for both sides – because every AI project has two.

Key Points at a Glance

• Co-determination: If an AI system can monitor performance or behavior, the works council has a co-determination right. What counts is suitability, not intent (Section 87(1) No. 6 BetrVG).
• Information and consultation: The employer must inform and consult the works council as early as the planning stage, expressly including the use of AI (Section 90 BetrVG).
• Expert: If the works council has to assess AI, calling in an expert is deemed necessary; an agreement with the employer is still required (Section 80(3) BetrVG).
• EU AI Act: From 2 December 2027, employers must inform the workers' representatives and the affected workers before using high-risk AI under Annex III, for example in HR. For systems already on the market or in service before then, this applies only after a significant change.
• Timing: If you involve the works council before the tool has been chosen, you will not end up negotiating over a mere yes or no – and through the works council, you bring the workforce on board as well.

Does the Works Council Slow Down AI? What the Numbers Say

The assumption persists: where there is a works council, AI makes only slow progress. The data do not support this. In 2025, the IAB surveyed around 15,000 establishments; one in four now uses generative AI. According to the analysis, which takes into account establishment size and industry, among other factors, there is no statistically significant relationship between the existence of a works council and the use of generative AI [1].

Another finding from the same study is revealing: only 27 percent of the establishments that use generative AI offer their employees training on it. A further 21 percent plan to, while 52 percent neither offer it nor plan to [1]. Training is precisely an area where the works council has rights of its own (Sections 96 to 98 BetrVG) – and here employers and works councils want the same thing: that AI tools are used well.

The Weizenbaum Institute offers a clue as to where things get stuck. In 224 interviews, works councils from establishments with 50 or more employees in selected industries reported how they are involved in decisions on new technologies [2]:

So at the planning stage, fewer than half of the works councils are regularly involved. The survey concerns new technologies in general, not just AI, and is not representative. But the numbers suggest that the works council often only comes into play once the important decisions have already been made.

What Rights Does the Works Council Have When It Comes to AI?

Since the Works Council Modernization Act of 2021, the German Works Constitution Act (BetrVG) has explicitly mentioned artificial intelligence [3]. Which rights apply depends on the phase of the project and what the system does:

Since 2021, it has been easier to bring in an expert: if the works council has to assess the introduction or use of AI, calling in an expert is deemed necessary. However, it still requires a more detailed agreement with the employer (Section 80(3) sentence 1 BetrVG) [3]. An expert whom both sides trust can clarify questions before they turn into disputes.

When Does AI Require Co-Determination? (Section 87 BetrVG)

The most important lever is Section 87(1) No. 6 BetrVG. It covers technical systems "designed to monitor the behavior or performance of employees." That sounds like intent, but it is meant more broadly: according to established case law, it is sufficient that a system is objectively suitable for collecting or recording information about behavior or performance. The employer's intention is irrelevant. The Federal Labour Court confirmed this again in 2024: even a headset system that allows supervisors to listen in on conversations requires co-determination – even if nothing is recorded [4].

For AI, this means: many systems fall under it because they log who enters what or assign results to individual people. An AI assistant that stores inputs with names is suitable for monitoring behavior or performance as soon as the employer can access these logs – even if nobody intends to evaluate them.

A second detail from the same decision is worth noting: the competent body there was not the local works council but the central works council [4]. Anyone planning to use a system uniformly at several sites should clarify which body is responsible before the first meeting.

What Applies to ChatGPT and Copilot?

So far there is essentially one decision on this: in 2024, the Hamburg Labor Court found no co-determination right under Section 87(1) Nos. 1, 6 and 7 BetrVG when an employer allowed the use of ChatGPT and similar tools [5]. The facts were decisive: the tools ran in the browser, the employees created their own accounts and bore any costs themselves, and there was no obligation to use them. The employer had no access to the data held by the provider.

No general rule follows from this. The ruling was made in expedited interim proceedings and concerned a single case. If the employer wants to offer an AI solution itself, that solution would, according to a commentary on the decision, be subject to co-determination [6]. This is likely to apply to tools such as Microsoft 365 Copilot, which are provided via company accounts and generate usage data that can be attributed to individuals.

What the EU AI Act Changes for AI in the Workplace – and When

In addition to the BetrVG, the European AI Regulation, the EU AI Act, applies [7]. It has been in force since August 2024 and applies in stages. In July 2026, Regulation (EU) 2026/1744 postponed several deadlines [8]. Guides written before that therefore still state August 2026 for high-risk AI. As of October 2026, the following applies:

In HR, AI is in principle considered high-risk if it is used in recruitment and selection (from placing targeted job advertisements to evaluating candidates), is involved in decisions on working conditions, promotion or termination, allocates tasks based on behavior or personal traits, or monitors and evaluates performance and behavior (Annex III No. 4) [7]. For systems that were placed on the market or put into service before 2 December 2027, the obligations – including the information obligation under Art. 26(7) – only apply after a significant change in their design (Art. 111(2)). Systems intended for use by public authorities must in any case meet the requirements by 2 August 2030 at the latest [7][8]. What this means for HR work is shown on our page on HR and human resources.

Important for both sides: the AI Act does not replace co-determination. The rights under the BetrVG already apply today – regardless of whether a system is classified as high-risk AI. How to classify your AI systems is described in our service EU AI Act Readiness; what the postponement means for compliance programs is covered in our article on AI compliance consulting.

What Belongs in a Works Agreement on AI

A works agreement on artificial intelligence can be concluded for each system individually or as a framework works agreement on AI that sets out principles for all systems. The 2024 Federal Labour Court case provides an example: there, a central works agreement from 2018 governed the introduction and use of the company's IT systems; systems subject to co-determination were each to be added to an annex as a system arrangement [4]. However, neither side can compel the other to conclude such a framework agreement [9].

Typical topics, based on the structure of the Bitkom guide [9]:

• Objectives and principles: what AI is to be used for in the company and which guidelines apply.
• Subject matter and scope: which systems are covered in which areas.
• Introduction and modification of systems: how the works council is involved in new systems and significant changes, such as new functions or models.
• Employee rights: information, access and data protection.
• Training: how employees are trained to work with AI.
• Performance and behavior monitoring: whether and within what limits data on performance or behavior may be evaluated.
• Works council oversight rights: how the works council can check compliance.
• Conflict resolution: how disputes are resolved, for example with an expert or through the conciliation committee.

Two decisions show what matters here. First, a works agreement does not replace data protection: according to the Court of Justice of the European Union, it must also comply with the requirements of the GDPR, and courts may review this in full [10]. Second, the works agreement is the limit: when testing HR software, an employer had transferred more real employee data than the works agreement allowed. The Federal Labour Court awarded an employee 200 euros in damages for the loss of control over their data [11]. That is why testing is done with test data wherever it is sufficient. What else data protection requires in HR is described in our article on data privacy in the HR department.

Good starting points for your own texts are the model texts of IG Metall NRW from the employee perspective and the Bitkom guide from the employer perspective [12][9]. The Bitkom guide reflects the status of February 2026: it still states the deadlines of the old version of the AI Act and mentions the postponement only as a proposal.

The Better Timeline: Four Steps to Agreement

The works council's rights cannot be planned away. But they can be built into the plan so that they support the project instead of holding it up. The difference lies in the timing: if the works council is only asked once everything has been decided, all it can say is yes or no. If it is involved from the start, it can help shape the project.

1. Bring both sides to the table – before the tool is chosen. In the first meeting, clarify what the employer side wants to achieve and what the works council side wants to protect. The meeting does not replace informing the works council under Section 90 BetrVG, but it makes that step easier.
2. Translate. Both sides often say the same thing and hear something different: one says "transparency," the other hears "control." Objections become conditions – data separated by purpose, access clarified, learning time planned.
3. A limited trial – agreed with the works council in advance. For example, 30 days with test data, a fixed end date, a clear data boundary and a joint evaluation date. If the system can monitor performance or behavior, the works council's co-determination right already applies to the trial.
4. Decide together. Continue, change or stop. Both sides say yes and know what they are agreeing to – and the agreement for regular operation adopts what has proven itself in the trial.

It helps to keep the works council's role in mind: it speaks for the workforce. If you bring it on board early, you bring the workforce along at the same time – and that is where it is decided whether an AI tool will actually be used or end up gathering dust. In line with this, the Weizenbaum Institute observes that where employers and works councils cooperate, employees are more involved in introducing AI, and both management and works councils rate employees' acceptance of AI somewhat higher [2].

What Works Councils Can Do Themselves

• Insist on early information: Under Section 90 BetrVG, the employer must inform the works council as early as the planning stage. Ask as soon as an AI project becomes known – not only when consent is requested.
• Ask about the classification: Does the system count as high-risk AI under the AI Act? Additional obligations from December 2027 depend on this.
• Help shape training: If AI changes tasks in such a way that employees' knowledge and skills are no longer sufficient for their work, the works council has a co-determination right when in-company training measures are introduced (Section 97(2) BetrVG) and in any case when they are carried out (Section 98 BetrVG).

Common Mistakes – and How to Avoid Them

• The Friday email: Demanding consent under time pressure provokes pushback. Build works council involvement into the project plan rather than leaving it until the end.
• Unclear responsibility: If you negotiate with the wrong body for a system used across several sites, you risk having to renegotiate later. Clarify early whether the local or the central works council is responsible.
• Leaving data protection to the works agreement: A works agreement does not replace the GDPR. Check both.

Is an AI project with a works council coming up? We translate between employer and works council so that both sides sit down together early on – more about this in our service AI Rollout with the Works Council. On 18 November, we will also talk about it at KI-X in Berlin. The fastest way is a direct conversation: book a 30-minute intro call.

Frequently Asked Questions About Works Councils and AI

Does the works council have to approve the introduction of AI?

If the system is suitable for monitoring the performance or behavior of employees, yes (Section 87(1) No. 6 BetrVG). This applies to many employer-provided AI tools, because they log inputs or assign results to individuals. Regardless of this, the employer must inform the works council as early as the planning stage (Section 90 BetrVG).

Does the works council have a veto on AI?

Not at the planning stage – Section 90 BetrVG gives it a right to information and consultation. Where it has co-determination rights, for example under Section 87(1) No. 6, the employer cannot introduce the system without agreement. If no agreement is reached, the conciliation committee decides.

Does the works council have to approve ChatGPT or Microsoft 365 Copilot?

It depends on the individual case. In 2024, the Hamburg Labor Court found no co-determination right under Section 87 BetrVG when employees used ChatGPT voluntarily in the browser with their own accounts. If the employer provides an AI tool itself, for example Microsoft 365 Copilot via company accounts, there are strong arguments for a co-determination right.

What belongs in a works agreement on AI?

Typical topics are objectives and principles, scope, how new systems and changes are handled, limits on performance and behavior monitoring, employee rights, training, the works council's oversight rights and conflict resolution. Principles can be set out in a framework agreement, with individual systems added later.

Who is responsible: the works council or the central works council?

As a rule, the local works council. The central works council is responsible if uniform rules for several establishments are strictly necessary – in the 2024 Federal Labour Court case, for example, because the system was administered centrally. The mere wish for a uniform solution is not enough. A local works council can, however, delegate a matter to the central works council (Section 50(2) BetrVG).

What applies in companies without a works council?

In that case, the participation rights of a local works council do not apply. If the company has a central works council that is responsible for a system used across several sites, its responsibility also extends to establishments without a works council (Section 50(1) BetrVG). However, the employer must inform the employees themselves about planned technical changes and their effects on the workplace (Section 81(4) BetrVG). Data protection applies in any case, and from 2 December 2027, employers must in principle also inform the affected workers before using high-risk AI under Annex III (Art. 26(7) AI Act); for systems already on the market or in service before then, only after a significant change (Art. 111(2) AI Act).

When do the obligations of the EU AI Act apply to AI in HR?

For high-risk AI under Annex III, for example AI used in hiring, promotion or performance evaluation, from 2 December 2027; for systems already on the market or in service before then, only after a significant change, and for systems intended for public authorities from 2 August 2030 at the latest. Prohibitions, such as the ban on emotion recognition in the workplace, have applied since February 2025, as has employers' obligation to support the development of AI literacy among their staff (Art. 4, amended as of 27 July 2026).

Conclusion

Whether the works council slows down an AI project or helps steer it is rarely decided by the law and often by the timeline. The rights are clear: information at the planning stage, co-determination for anything that can monitor performance or behavior, and additional information obligations under the AI Act from December 2027 if a system counts as high-risk AI. If you bring the works council to the table early, you will not end up negotiating over a yes or no, but over a project that both sides have helped shape.

This article reflects the legal situation in Germany as of October 2026 and is no substitute for legal advice.

References

1. (IAB Brief Report 8/2026: Artificial intelligence in German establishments – one in four now uses generative AI (in German))
2. (Weizenbaum Institute, Discussion Paper 39: Two worlds of AI in the world of work (2024, in German))
3. (German Works Constitution Act (BetrVG, in German))
4. (Federal Labour Court, decision of 16 July 2024 – 1 ABR 16/23 (in German))
5. (Hamburg Labor Court, decision of 16 January 2024 – 24 BVGa 1/24, commentary by LTO (in German))
6. (KPMG Law: Hamburg Labor Court – permitting ChatGPT is not subject to co-determination (2024, in German))
7. (Regulation (EU) 2024/1689 (AI Act), consolidated version of 27 July 2026)
8. (Regulation (EU) 2026/1744 amending the AI Act)
9. (Bitkom: Guide on artificial intelligence and co-determination (status 02/2026, still states the deadlines before the AI Act amendment; PDF, in German))
10. (Court of Justice of the EU, judgment of 19 December 2024 – C-65/23)
11. (Federal Labour Court, judgment of 8 May 2025 – 8 AZR 209/21 (in German))
12. (IG Metall NRW and TBS NRW: Works agreements for artificial intelligence made easy – model texts (2025, PDF, in German))