The legal basis
Section 87 (1) no. 6 of the Works Constitution Act (BetrVG) covers technical systems that are intended to monitor the conduct or performance of employees. The Federal Labour Court (Bundesarbeitsgericht) has read the word "intended" broadly for decades: it is enough that the system is objectively capable of enabling monitoring, regardless of its original purpose and regardless of whether the employer actually wants to analyse the data.
This is not a special rule for AI. It is the same provision that applies to time recording, telephone systems and software logs. What is new is only that AI tools process and condense personnel data at a depth earlier systems did not reach. A language model that proposes the five best matches from twenty applications is evaluating people. That is monitoring in the sense of the provision, even if nobody in the HR team would call it that. For readers used to UK or US practice, the closest comparison is this: there is no equivalent of an enforceable veto in those systems, and that is exactly why the German rule catches international groups off guard.
The three criteria
If all three apply, you need a works council agreement before the tool goes into productive use. How it comes together: The Works Council AI Agreement. What applies until then: An AI policy for HR teams.
What falls under it, and what does not
The question is not whether a tool is called AI, but what happens to personnel data. Six typical cases:
- Application screening: yesThe tool evaluates people. Subject to co-determination and high-risk under the EU AI Act.
- Reference letter drafted from bullet points: yesPerformance data goes in, and the wording is an evaluation.
- Summary of appraisal meetings: yesConduct and performance data, condensed. Clear-cut.
- HR chatbot for policy questions: yesAs soon as it knows who is asking and logs the questions.
- Job ad without personal data: usually noAn ad text evaluates nobody. Careful once applications are matched against it.
- Microsoft Copilot in Outlook: it dependsOn general email: no. On the HR inbox with applications: yes.
What happens if the works council is bypassed
If the works council is bypassed, it can have the tool stopped through an injunction, obtained within days. Everything produced with the tool until then, for instance pre-selection decisions on applications, stands on shaky legal ground and can be challenged in a dispute. A rejected candidate who learns that a tool without co-determination screened her out has an argument that holds up in court.
The second cost is quieter and higher. A works council that has been bypassed once negotiates the next agreement paragraph by paragraph. That costs months during which the tool stands idle. Involve the works council before the start and it is on your side from the second step onward. In my mandates, the works councils that pushed back hardest were the ones that found out about a tool from a colleague rather than from HR.
What the works council can demand
The works council can demand that the introduction is halted until an agreement is in place. It can demand insight into how the system works, which for AI tools means the data basis, the prompts and the logs. It can bring in an expert whose costs the employer bears. And it can enforce conditions: which use cases are permitted, which are prohibited, how long data is kept, who makes the final decision. What it cannot do is decide unilaterally. If no agreement is reached, a conciliation committee (Einigungsstelle, a joint body with a neutral chair) decides, not the works council and not the employer.
What it means in practice
Before an AI tool goes into productive use in the HR team, ask whether it processes personal data, whether it could evaluate conduct or performance, and whether several people use it. If all three apply, a works council agreement on AI is needed before go-live. The practical sequence: use-case inventory with the works council, risk assessment, drafting, test phase. Two to four months, with the routine drafts prepared in parallel so they go live the day the agreement is signed. What that means for the individual routines: AI Screening Criteria, AI for Reference Letters, AI for Employee Questions.
Co-determination on system rollouts and AI routines with works councils Roles in corporate groups, mid-sized companies and scale-ups. Practical guidance, not legal advice. More about me
Checklist: is your AI tool subject to co-determination?
The three criteria from Section 87 (1) no. 6 BetrVG as a short checklist for every new tool, with the six typical cases and a note on what to do next. One page, for the next meeting with management or the works council.
Frequently asked questions
Does Section 87 apply to small companies without a works council?
No. The co-determination right presupposes an existing works council. Without one, data protection and the EU AI Act still apply, and an AI policy is the right instrument.
Is consulting the works council enough, instead of an agreement?
No. Section 87 (1) no. 6 is an enforceable co-determination right, not a mere duty to consult. Without agreement, a conciliation committee decides, not the employer.
What if the tool only drafts text and does not analyse data?
Then the question is whether it is capable of monitoring. A plain job ad text without personal data usually falls outside. As soon as applications, reference letters or performance data go in, the capability is there.
Can the works council ban an AI tool outright?
It can refuse the introduction, and then the conciliation committee decides. In practice it is rarely about a ban. It is about conditions: which use cases, which data, which monitoring rights.
Does this apply to tools mandated by the parent company?
Yes. The co-determination right applies in the German establishment regardless of who selected the tool. International subsidiaries underestimate this regularly.