Starting August 2, 2026, businesses across the European Union must clearly label any content generated by artificial intelligence — or risk hefty penalties. Article 50 of the EU AI Act, which takes effect that day, requires providers and users of AI systems to disclose when people are interacting with chatbots and to mark AI-produced texts, images, and deepfakes.
The European Commission published a guidance document on July 20, 2026, spelling out the transparency obligations and a related code of conduct. A separate code for labeling AI-generated content had already been released in June. Companies that fail to comply face fines of up to €15 million or 3 percent of their global annual turnover, whichever is higher.
To determine whether something qualifies as a deepfake,
Tech Giants Sign On, but Industry Voices Concern
Google signed the EU’s labeling code of conduct on July 24, 2026, committing to use the C2PA standard and its own SynthID watermarking technology. The company joined Apple, ElevenLabs, Kakao, NVIDIA, and OpenAI in the initiative.
Yet not everyone is celebrating. Karen Massin of Google warned on July 23, 2026, that overregulation could backfire. “Additional regulatory complexity may confuse consumers and harm Europe’s competitiveness,” she said. The German Chamber of Commerce and Industry (DIHK) echoed that sentiment, noting that many businesses remain uncertain because detailed guidelines were published only days before the rules took effect.
Germany Moves on Enforcement and Liability
On July 23, 2026, Germany cleared the path for a national implementation law after the Bundesrat gave its approval. The Federal Network Agency (Bundesnetzagentur) will serve as the central market surveillance authority for AI systems. The TÜV association urged authorities to quickly establish the necessary supervision, accreditation, and notification processes.
That same day, the Munich I Regional Court ruled that platform operators are liable for AI-generated summaries — a decision that clarifies legal responsibility in the new landscape.
Training Requirements and Delayed Deadlines for High-Risk AI
Under Article 4 of the AI Act, authorities can demand proof of AI competence from companies starting in August. Several providers have already launched training solutions. On July 23, 2026, Provimedia GmbH introduced an online course available in eight languages. It takes 60 to 90 minutes to complete and covers ten chapters, with a mandatory section of five chapters containing 46 learning cards and 18 exam questions. The verifiable certificate is valid for one year.
Also on July 23, Fraunhofer FIT unveiled “Kassandra,” an AI exit game designed to teach users about risk classes and how to critically evaluate AI outputs through gameplay.
While the transparency rules take effect now, deadlines for high-risk AI systems have shifted. The “Digital Omnibus on AI,” published July 24, 2026, and entering force July 27, 2026, pushes those deadlines to December 2, 2027, for standalone systems and August 2, 2028, for embedded systems.
