Since August 2, 2026, the European AI Regulation (Regulation 2024/1689) has been in full effect. This means German companies must now comply with far-reaching rules that severely restrict the use of AI in the workplace – particularly for employee surveillance.
Key Facts
- Banned practices: Social scoring, predictive policing, and emotion recognition in the workplace are now prohibited (valid since February 2, 2025, now fully enforced)
- Fines: Violations cost up to €35 million or 7% of global annual turnover
- Market surveillance: The German Federal Network Agency (Bundesnetzagentur) enforces compliance in Germany
- AI competence: Companies must take measures since July 27, 2026 to ensure decision-makers and users understand how AI systems work and their risks
What Is Now Banned?
The regulation prohibits three central surveillance practices: social scoring (automatic employee ratings), predictive policing (predicting misconduct), and the use of emotion recognition – AI systems that analyze employee emotions. These rules aim to protect worker rights and prevent abusive surveillance.
In Germany, the Federal Network Agency oversees enforcement under the AI Model Implementation Act (KI-MIG). Companies that violate these bans face severe penalties: fines up to €35 million or 7% of global annual turnover.
New Requirement: AI Competence in Organizations
A second major pillar concerns AI competence. Since July 27, 2026, companies must take steps to ensure that decision-makers and users understand how their AI systems function and what risks they pose. This can be achieved through training, documentation, or internal guidelines.
However, there is a German peculiarity: violations of these competence requirements currently carry no fine. Experts refer to this as an enforcement gap in national law that could complicate compliance.
AI-Generated Content Labeling Becomes Mandatory
Another central aspect is the machine-readable labeling of AI-generated content. The EU requires such content to be clearly identifiable as artificially created. To meet these requirements, approximately 190 companies have signed an EU code of conduct – including industry leaders like Google, Meta, Microsoft, Mistral, and OpenAI.
Anthropic announced that its Claude models have been equipped with invisible watermarks and special C2PA metadata for images since early August 2026. These labels are designed to persist even after copying or partial editing.
What Does This Mean for German Employers?
The new legal framework presents concrete challenges for companies: those using AI systems for employee surveillance must review and potentially discontinue them immediately. Simultaneously, it makes sense to establish AI competence training programs now – not just for compliance, but to understand risks. The enforcement gap regarding AI competence could change, so proactive measures are advisable. Works councils also have co-determination rights for AI systems used to monitor employees – this must be considered in companies with employee representation.
Sources
Editorially owned by Ideal Syka. Sources and method: Newsroom & method. Tips and corrections: ai@i6eal.de.
