- The EU AI Act’s Article 50 transparency rules took effect on August 2, 2026, requiring any interactive AI system to disclose its non-human status and any AI-generated deepfake content to carry visible labels and machine-readable watermarks.
- Germany’s Federal Network Agency (Bundesnetzagentur) was formally activated as the central AI market surveillance authority when the KI-MIG law entered into force on July 29, 2026 — just days ahead of the Article 50 deadline, after missing the EU’s original August 2, 2025 designation deadline by nearly a year.
- Non-compliance with the Article 50 obligations carries fines of up to €15 million or 3% of global annual turnover, and the rules apply to any AI system covered by Article 50, not only high-risk ones, meaning the compliance burden falls on a broad range of businesses using generative AI.
Germany’s enforcement law is already overdue, and the EU’s new AI transparency rules are now in force anyway. On August 2, 2026, Article 50 of the EU AI Act became applicable, requiring providers and deployers of interactive AI systems to tell users they are talking to a machine, and mandating visible labels on AI-generated deepfake content. The rules carry fines of up to €15 million or 3% of global annual turnover for non-compliance.
What Article 50 Requires
The transparency obligations centre on two distinct duties. Interactive AI systems, chatbots, AI agents and similar tools, must explicitly inform users that they are not communicating with a human. The aim is to prevent deception at the point of engagement rather than through a buried terms-of-service disclosure.
The second duty covers AI-generated or AI-manipulated content that depicts real persons, objects, places or events: what the Act refers to under the deepfakes category. Providers must ensure such content carries both a clear visible label and a machine-readable mark. The European Commission has published a set of standardised icons for this purpose, alongside guidelines to help businesses implement the requirements. A Code of Practice on AI-generated content has been signed by more than 180 organisations, according to the Commission.
One timing nuance matters for compliance teams. Generative AI systems already on the market before August 2, 2026 have until December 2, 2026 to implement machine-readable watermarking under Article 50(2). The labelling obligation applies from today; the technical watermarking standard for legacy systems gets an additional four months.
Germany’s Enforcement Gap
Member states are responsible for building their own enforcement frameworks under the Act. Germany has designated the Federal Network Agency (Bundesnetzagentur) as its central market surveillance authority for AI, a move formalised under the AI Market Surveillance and Innovation Act, known by its German abbreviation KI-MIG. The KI-MIG also establishes an AI Coordination and Competence Centre to manage cooperation between domestic authorities and serve as the interface with EU-level bodies, and it provides for AI regulatory sandboxes to allow startups and enterprises to test compliant systems in real-world conditions.
There is a structural problem, though. The KI-MIG is still a draft law awaiting debate in the Bundestag. Its own deadline for enactment was August 2, 2025, a date that has already passed. Germany is enforcing EU-level rules with national legislation that is not yet on the books.
Sector-specific oversight sits with existing regulators in the interim: the Federal Financial Supervisory Authority (BaFin) covers AI in financial services, while state-level authorities handle AI used by regional government bodies. How well that patchwork holds until the KI-MIG is enacted remains an open question.
Where the Phased Rollout Stands
The EU AI Act’s staggered compliance timeline means different obligations are live at different times. Prohibitions on unacceptable-risk AI practices, government social scoring among them, took effect on February 2, 2025. Rules for general-purpose AI models, covering technical documentation, copyright policy implementation and training-data summaries, became applicable on August 2, 2025.
The most demanding tier, covering high-risk AI systems with potential to affect health, safety or fundamental rights, has been pushed back. A provisional “Digital Omnibus” agreement reached in May 2026 proposes deferring obligations for standalone Annex III systems to December 2, 2027, and for high-risk AI embedded in regulated products under Annex I to August 2, 2028.
The Article 50 transparency obligations now in force sit outside that high-risk tier entirely. They apply to any AI system used in the situations Article 50 covers, regardless of risk classification. That scope is wider than many compliance teams initially anticipated: a business using generative AI for customer-facing content or interactive services falls within the rules even if its systems would never be classified as high-risk. The Bundesnetzagentur and its counterparts across the EU are now responsible for enforcing that.
