AI and the Works Council: 5 Steps to Legally Sound AI.
Bring the works council in early, and come with clear guardrails. That's how you unite innovation and co-determination in your company.
30 minutes, online, free and non-binding.
AI rollouts rarely fail because of the technology, they fail due to a lack of legal certainty and internal resistance. The roadmap rests on three pillars, local hardware, technically enforced rules, and enablement, and moves through five phases: from early works council involvement, through foundation, anonymization, and pilot, to a trained rollout.
Uniting innovation and co-determination
You've recognized the urgency. You know you need to automate. But how do you actually start, without sinking into a swamp of legal concerns and resistance?
This isn't a theoretical paper, it's your operational roadmap. It translates the "sovereignty doctrine" into five phases, so AI projects are not just planned but actually put into live operation. The most important difference from many failed projects: the works council sits at the table from day one.
Bring the works council in early, and come with clear guardrails. Security first, then innovation.
The 3 pillars of your strategy
- Technology: local hardware instead of the public cloud, as a shield against data flowing to insecure third countries.
- Governance: automated rules instead of manual control. Compliance can't depend on good intentions, it has to be architecturally enforced.
- People: enablement instead of replacement. Your employees learn to master the technology, to ease concerns and help shape the change themselves.
What the law requires
- Notification: the employer must inform the works council about the planning of work procedures and processes, explicitly including the use of AI, in good time, and consult with it on the plan (Section 90 of the German Works Constitution Act, BetrVG). "In good time" means while there's still something to decide.
- Co-determination: if an AI system is capable of monitoring employee behavior or performance, it requires the works council's consent (Section 87(1) No. 6 BetrVG). This applies to most office tools, since what matters is the capability, not the intent.
- Experts: if the works council needs to assess the introduction or use of AI, bringing in an expert is considered necessary (Section 80(3) BetrVG).
- Selection guidelines: if AI is used in hiring, transfers, or dismissals, co-determination on selection guidelines applies (Section 95(2a) BetrVG).
- EU AI Act: employers deploying a high-risk AI system in the workplace must inform employee representatives and affected employees in advance (Art. 26(7) of the EU AI Act, from December 2, 2027 under the Digital Omnibus). The obligation around AI literacy (Art. 4) has already applied since February 2, 2025.
The 5-phase implementation plan
Phase 1: The dialogue (bring the works council in early)
Inform the works council as soon as you're planning an AI initiative, not once the technology is already in place. Bring goals, planned use cases, and initial guardrails to the table, and work out the rest together.
- Disclose goals and use cases: what should the AI do, and what explicitly should it not do?
- Agree on a "negative list", for example: no performance or behavior monitoring.
- Kick off a framework works agreement on AI: purpose, data types, retention periods, analysis bans, change procedures.
- Define a clear escalation path for when rules are broken.
Phase 2: The foundation (hardware and setup)
Build the technical foundation for a local AI instance, exactly as agreed in Phase 1. Involve your data protection officer.
- Procure local hardware (GPU servers) or secure dedicated hosting in Germany.
- Make sure no internet connection to third-party providers is required.
- Isolate sensitive networks and, where needed, carry out a data protection impact assessment.
Phase 3: The "digital gatekeeper" (configuration)
Set up the anonymization layer. It's your key technical argument against surveillance and privacy concerns.
- Set up PII masking: names, for example, are replaced with the placeholder PERSON.
- Configure the "shredder": the system clears short-term memory after every session.
- Test the masking with dummy data and demonstrate it live to the works council ("glass box").
Phase 4: The sandbox (piloting)
Start small in a controlled environment, within the agreed rules, and avoid an unwieldy "big bang" rollout.
- Choose an innovation-friendly team as the pilot group, on a voluntary basis.
- Define three concrete use cases, for example email drafting or meeting summary.
- Regular feedback with the pilot group and works council: where are the sticking points? Where do concerns remain?
Phase 5: The license (training and rollout)
Only release the tools once employees are trained. Technology without competence just creates fear, and AI literacy is a legal requirement under Art. 4 of the EU AI Act anyway.
- Training through the internal academy.
- Internal proof of training completion ("AI license").
- Go-live for the whole company based on the works agreement.
In a free initial call, we will look together at what matters most for you.
Checklist: are you "audit-ready"?
The EU AI Act, GDPR, and the BetrVG demand diligence. These five points will get you prepared:
- Works council informed in good time and a works agreement concluded.
- Local data storage: prevents data transfer to the US and lowers GDPR risk.
- Anonymization (proxy): defuses the accusation of employee surveillance.
- Audit logs (WORM): prove, in case of dispute, that no manipulation or performance review took place.
- Employee training: fulfills the AI literacy obligation under Art. 4 of the EU AI Act.
Without a roadmap, you risk endless discussions, high legal fees, unsettled employees, and an IT department left chasing shadow IT like covert ChatGPT use. With a roadmap, a structured process emerges where compliance becomes an enabler.
As of October 29, 2026. This guide does not replace legal advice. Works agreements should be coordinated with your works council and, where appropriate, your legal counsel.

Download as PDF.
Leitfaden as a PDF, 5 pages. Enter your details below to start the download right here.
Before you reach out
What's the best way to involve the works council?
Early and openly: inform them as soon as you're planning (Section 90 BetrVG), and work out guardrails together. Later, a demonstration tends to win people over: once the works council sees how the digital gatekeeper masks names, surveillance concerns are usually easy to resolve.
Is a local AI ("on-premise") strictly necessary?
For maximum sovereignty and strict GDPR standards in core processes, on-premise is the safest path. It rules out US data transfers and ends discussions about the CLOUD Act.
What does implementing the 5 phases cost?
That depends on company size and the use cases chosen. We estimate the effort after the initial consultation. The phases can be implemented as a compact pilot project. Our overview of Digitalization Funding Programs 2026 shows available funding options.
How does the digital gatekeeper work?
It's a rule-based intermediary layer. Before your text goes to the language model, the gatekeeper filters out personal data (PII) and replaces it with placeholders. The AI never sees the real names of your customers or employees.
Start phase 1 with a conversation.
Let's check together whether your infrastructure is ready for sovereign AI. We'll get back to you within 24 hours with a proposal for the initial consultation.

