There is a predictable point where enterprise AI initiatives in co-determined organizations come to a grinding halt, and it has nothing to do with technical feasibility. Contracts are signed, licenses are paid, the pilot department is eager to start, and then the works council steps in and requests an indefinite pause to review the system first.
From that moment on, management negotiates from the weakest possible position: capital has been committed, rollouts are scheduled on executive calendars, and you have zero immediate alternatives. The works council is fully aware of this leverage. What follows is months of drawn-out dispute, often ending in a restrictive company agreement that satisfies no one.
The frustrating reality is that this friction rarely stems from ideological opposition. It is almost entirely a matter of sequencing.
The Legal Framework in Three Sentences
The central co-determination right under German labor law is governed by § 87 (1) No. 6 of the German Works Constitution Act (BetrVG): the works council has mandatory co-determination rights regarding the introduction and application of technical equipment intended to monitor employee behavior or performance. Under well-established case law of the Federal Labour Court (BAG), the objective suitability for monitoring is entirely sufficient; employer intent is explicitly irrelevant (BAG, Decision of 25 September 2012, 1 ABR 45/11).
This point is regularly underestimated. Any AI system that produces analytics, automated ratings, or performance rankings is objectively suitable, even if no one intends to use it for employee surveillance. An assistant logging handling times falls into the exact same category. You do not need to intend surveillance; the mere technical possibility is enough.
And because this co-determination right is legally enforceable, deploying a tool without prior agreement invites injunction claims, mandatory rollbacks, and legal disputes over whether collected data can even be utilized. Additionally, the statutory duty to provide information under § 90 BetrVG requires timeliness, which case law defines as consultation prior to irreversible capital commitments, not after. That is precisely where the critical mistake happens.
When AI tools influence hiring, promotions, or staff selection, formal selection guidelines under § 95 BetrVG apply as well. And under the EU AI Act, employers must inform worker representatives before putting high-risk systems into operation.
Private Employee Usage Is the Exception, Not the Standard
In 2024, the Hamburg Labor Court ruled that purely personal use of tools like ChatGPT by individual employees on private accounts, without integration into corporate IT systems and without management directives, does not trigger enforceable co-determination under § 87, because the employer exercises no direct technical control over the system.
This decision is often cited as a reassuring green light. It is anything but. The ruling was exceptionally narrow and collapses the moment an employer licenses the tool, funds subscriptions, mandates its use, or integrates it into internal infrastructure, precisely the moment informal experimentation turns into an enterprise initiative.
Relying on this judgment means accepting that employees use unmonitored private accounts for business tasks. That may be legally convenient in the short term, but it represents the far more hazardous problem for data security and trade secrets.
Why Negotiations Drag On
In practice, these meetings typically bring together two parties that are both missing the information a decision would rest on. Executive leadership can rarely specify what exact data the tool processes, where telemetry is stored, or how training pipelines use inputs. The works council, in turn, cannot reliably distinguish actual employee hazards from harmless software features and defaults to obstruction, which is their statutory duty when solid facts are missing.
The typical outcome is a rigid company agreement that either bans anything not explicitly enumerated or remains so vaguely phrased that it proves useless upon the first dispute. Notably, works councils have the legal right to retain external experts when evaluating AI systems, an entitlement reinforced in recent years that remains underutilized.
The Sequence That Actually Works
Involve the works council before selecting a vendor, not after. That sounds like relinquishing control, but it is the exact opposite: as long as nothing is decided, you negotiate criteria rather than defending a finished product purchase. And you retain a vital piece of bargaining leverage you will lose later: the credible option to choose an alternative provider.
State what you want to achieve, not which tool you want to buy. “We want to accelerate quotation workflows” is open to constructive negotiation. “We are introducing Copilot” is an announcement to which there are only two responses.
Address the monitoring issue proactively before it is raised. Clarify what logs are generated, who can access them, retention periods, and what explicitly will never be done with them. A voluntary self-commitment at this stage costs you virtually nothing and disarms the other side’s strongest argument.
Framework agreement rather than case-by-case rules. If you renegotiate for every individual software tool, you will be negotiating continuously from now on. A framework works agreement on AI usage with a standardized procedure for onboarding new systems is the only way to streamline rollouts. In early 2026, Bitkom published the guideline „AI and Co-Determination – Cornerstones of a Works Agreement for the Use of AI in Enterprises“, detailing standard regulatory scope: applicability, introduction process, employee rights, training, performance monitoring, and audit rights. That provides a practical blueprint for your first draft.
Train both sides together. A negotiation over a system whose workings are unclear on both sides will never yield a good outcome. Investing half a day to build a shared foundational understanding is the most cost-effective line item in the entire project.
What Lies Ahead
Regulatory oversight will continue to expand. The European Commission is already preparing directives regarding algorithmic management in the workplace. Companies with a robust framework agreement in place today will simply update their guidelines; those without one will be forced to start from scratch under severe time pressure.
The golden rule from my operational experience: a company agreement is the prerequisite for an AI rollout, not its concluding milestone. Putting it first costs four to eight weeks of planning; putting it last can cost the entire project.
This article does not constitute legal advice. Drafting formal company agreements requires specialized employment law counsel.
Sources
- § 87 German Works Constitution Act (BetrVG) – Co-Determination Rights – Federal Ministry of Justice
- Bitkom Guidelines: “AI and Co-Determination” (February 2026)
- Federal Labour Court (BAG), Decision of 25 September 2012, 1 ABR 45/11 – on the objective suitability of technical systems
- Hamburg Labour Court (ArbG Hamburg), 2024 decision on private employee use of generative AI
- Regulation (EU) 2024/1689 (EU AI Act), Article 26 – Information obligations toward worker representatives
How this article was written: Grounded in statutory text, the Bitkom framework, and practical experience guiding introduction projects in co-determined enterprises.
About the Author
Philip Hohn advises medium-sized enterprises on deploying artificial intelligence. He is Managing Director of HBC Hohn Business Consulting UG and runs several of his own companies and projects, including Edura Akademie GmbH, an AZAV-accredited vocational training provider for AI enablement. Previously, he led an agency with 70 employees across four locations as well as a software development firm with fifteen engineers. He is not an attorney; legal observations in his texts do not constitute legal counsel in individual cases.