Introducing AI in the Hospital: Why the Betriebsrat Has a Say

The project is fully funded, the vendor selected, the pilot due to start in four weeks. Then a letter from the Betriebsrat — the works council — lands in the medical director's inbox: they learned of the project from the staff newsletter and request a pause until co-determination has been clarified. IT is irritated, the vendor waits, the timeline tips over.

The episode is avoidable — with less effort than the project team believes, and considerably earlier than is customary. For the legal situation is clearer than its perception in many project plans: with AI tools in hospital operations, co-determination is the normal case.

§87 (1) no. 6: the norm that almost always applies

The Betriebsrat co-determines in the "introduction and use of technical devices designed to monitor the behavior or performance of employees" (§87 (1) no. 6 of the BetrVG, Germany's Works Constitution Act) [1]. That sounds like time clocks and video cameras. The settled case law of the Bundesarbeitsgericht, Germany's Federal Labor Court, reads it more broadly: it suffices that the device is objectively suitable for monitoring; an intention to monitor does not matter [2]. An AI system that logs who generated and edited which draft when — and practically all of them do — is objectively suitable. For AI provided by the employer, co-determination is thus regularly triggered.

The most prominent court decision on generative AI in the workplace so far confirms this by converse implication. The Hamburg labor court in 2024 saw no co-determination right when an employer permitted the use of ChatGPT via employees' private accounts — precisely because the employer had no access there to behavioral or performance data (decision of 16 January 2024 – 24 BVGa 1/24) [3]. If the hospital provides the tool itself, with its own accounts and its own logs, the case is different. How to deal with unregulated private use is covered in the article on shadow AI in the department (in German).

Involvement begins before the selection

Since the 2021 amendment, the statute names artificial intelligence expressly: when AI deployment is being planned, the Betriebsrat must be informed in good time and the planned measures discussed with it (§90 (1) no. 3 BetrVG); if it brings in an expert on AI questions, that expert's necessity is deemed established (§80 (3) BetrVG) [4]. The legislature deliberately moved the involvement forward — before the selection decision, not after the contract signing.

If no agreement is reached, the Einigungsstelle — the statutory conciliation board — decides; its ruling replaces the agreement between employer and works council (§87 (2) BetrVG) [1]. How long such proceedings typically take for AI introductions is not reliably documented. What is documented is the risk of the shortcut: an introduction without involvement can be stopped by the Betriebsrat through an injunction — and then the project stands where the scene at the start of this text ends, only later and more expensively.

What belongs in the works agreement

The good news: the content of an AI Betriebsvereinbarung — a works agreement — is by now no uncharted territory. From negotiation practice, as documented for instance in the Bitkom guide "KI und Mitbestimmung" (February 2026), recurring building blocks have emerged [5]:

• purpose limitation and the express exclusion of performance and behavioral monitoring
• approved tools and use cases, with an annex mechanism for new tools and updates — otherwise negotiation starts over with every version
• handling of personal data: storage locations, deletion periods, logging, prohibitions on analysis
• training, congruent with the AI literacy duty of Article 4 of the EU AI Act
• the Einigungsstelle as the defined conflict mechanism

Anonymity you can demonstrate

The toughest point of almost every negotiation is mistrust of analyzability: what can the employer read out of the system about individuals — today, and after the next update? Assurances in text form ("we do not analyze at the level of individuals") are the usual route. Negotiations run recognizably differently when the assurance is built into the architecture and can be demonstrated: then the works council examines a property, not a promise.

One example of what that can look like concretely is the survey suite by aiomics — in pilot operation at rehabilitation clinics; the employee survey package is part of the planned expansion. There, the anonymity floor is built into the code: the system does not deliver analyses below ten responses, enforced at query level and complemented by complementary cell suppression; the employee channel stores neither person identifiers nor IP addresses, and timestamps are coarsened. A works agreement template and a technical anonymity whitepaper are part of the package as product artifacts, and activating the employee survey is technically blocked without a documented works council review — designed so that the Betriebsrat can verify the assurances itself. More on the suite in the article on PROMs and PREMs (in German).

The sequence that works

1. Inform before the selection is made (§90 BetrVG). A works council that learns of the project from the staff newsletter negotiates differently from one that helped formulate the requirements.
2. Settle the expert question proactively (§80 (3) BetrVG) — it comes anyway, and an expert named early tends to accelerate rather than slow things down.
3. Aim for a framework works agreement with an annex mechanism instead of a separate agreement per tool.
4. Demand demonstrables from the vendor: which person-level analyses are technically possible? The best answer can be shown — in the system, not in the contract annex.

If you are facing an AI introduction and want to structure the works council perspective early, write to us — the conversation is worthwhile even without a product angle. Ongoing analysis of AI, law and hospital operations comes in our weekly briefing Visite (German; English edition Grand Rounds is in preparation).

Sources

1. §87 BetrVG — Mitbestimmungsrechte. https://www.gesetze-im-internet.de/betrvg/__87.html
2. Zur objektiven Eignung als Maßstab der ständigen BAG-Rechtsprechung bei §87 Abs. 1 Nr. 6 BetrVG: Wilke Witte, „§87 Abs. 1 Nr. 6 BetrVG und die Mitbestimmung bei KI-Systemen“; Dr. Datenschutz, „Künstliche Intelligenz und Mitbestimmungsrechte des Betriebsrats“. Fachbeiträge. https://www.wilkewitte.de/post/87-abs-1-nr-6-betrvg-und-die-mitbestimmung-bei-ki-systemen ; https://www.dr-datenschutz.de/kuenstliche-intelligenz-und-mitbestimmungsrechte-des-betriebsrats/
3. ArbG Hamburg, Beschluss vom 16.01.2024 – 24 BVGa 1/24. Besprechung: Legal Tribune Online. https://www.lto.de/recht/hintergruende/h/arbg-hamburg-24bvga-1-24-chatgpt-einsatz-arbeit-beteiligung-betriebsrat
4. §90 Abs. 1 Nr. 3, §80 Abs. 3 BetrVG (i. d. F. des Betriebsrätemodernisierungsgesetzes 2021). https://www.gesetze-im-internet.de/betrvg/
5. Bitkom, Leitfaden „Künstliche Intelligenz und Mitbestimmung“, Februar 2026. https://www.bitkom.org/sites/main/files/2026-02/bitkom-leitfaden-kuenstliche-intelligenz-und-mitbestimmung.pdf

Sources accessed in July 2026.