AI at Work: When the Works Council Must Have a Say

In Germany, the works council must co-determine AI at work as soon as a tool can monitor employees' behavior or performance. This is set out in Section 87 (1) no. 6 of the Works Constitution Act (Betriebsverfassungsgesetz, BetrVG). The mere possibility of monitoring is enough — the tool does not have to actively watch anyone. Many AI assistants fall under this rule because they can capture inputs and usage. So involve the works council early, ideally before rollout.

The Short Answer: When Does the Works Council Get a Say?

There are two levels. First, the works council has a right to information and consultation when new technology is planned. Then, in many cases, a real co-determination right follows. Co-determination means you cannot introduce the tool without the council's approval. These points trigger involvement:

• The AI tool can measure, log, or evaluate behavior or performance.
• You introduce new software or a new work process.
• AI plays a role in selection decisions, such as hiring or dismissal.
• The use creates data about individual employees.

The Four Key BetrVG Provisions

Four sections decide the matter for AI at work. Two give the works council a strong co-determination right; two strengthen its position. The table below shows what each provision triggers:

Important: Section 87 (1) no. 6 applies whenever the technology is suitable for monitoring. It does not matter whether you actually intend to monitor anyone.

How to Involve the Works Council Early

Informing the works council late risks delay and conflict. An early, open process is faster and safer. Proceed in this order:

1. 1Inform the works council as soon as the AI plan becomes concrete (Section 90).
2. 2Disclose which data the tool processes and where it flows.
3. 3Offer a joint consultation and allow enough time.
4. 4Accept that the council may bring in an outside expert (Section 80 (3)).
5. 5Negotiate a works agreement before you roll out the tool.
6. 6Put rules, limits, and control rights in writing.

The Works Agreement on AI

A works agreement (Betriebsvereinbarung) is the usual way to settle co-determination cleanly. It creates clarity for both sides and prevents later conflict. Useful contents include:

• The purpose and permitted use cases of the AI, clearly bounded.
• Which data is captured and which is expressly not.
• A ban on covert performance and behavior monitoring.
• Storage periods, deletion deadlines, and access rights.
• Employee training and a named contact person.
• Rules for changes and a right to terminate the agreement.

GDPR and Employee Data

Alongside the BetrVG, the General Data Protection Regulation (GDPR) applies. Employee data is personal data and specially protected. Two principles matter here:

• Data minimisation (GDPR Art. 5): process only what is truly necessary.
• Purpose limitation: use data only for the agreed purpose.
• Transparency: employees must know what happens to their data.
• Note: pseudonymized data remains personal data (Recital 26). It still falls under the GDPR.

In short: technical safeguards replace neither co-determination nor data-protection duties. But they do reduce the risk that sensitive data reaches an AI tool at all.

How ONYRI Sanitize Reduces the Risk

ONYRI Sanitize detects sensitive data in a prompt and replaces it with reversible placeholders (tokens) before the text reaches an AI. After the answer comes back, the tool restores the original values. So the external model only ever sees anonymized text. A few points for context:

• The tokenizer runs 100% in the browser. The mapping from token to original value never leaves the browser.
• In the in-app Chat, only the already-anonymized text is sent to the model.
• This reduces data exposure and supports data minimisation under GDPR Art. 5.
• It is not a guarantee of GDPR compliance. Pseudonymized data stays personal, and the works council's right to co-determine remains.

So a tool like this does not replace a process. It complements the works agreement and data protection by letting less sensitive data leave your organization.