Europe is about to force AI to announce itself. With the EU AI Act’s transparency rules taking effect on August 2, 2026, the bloc has published a Code of Practice for labeling AI-generated content — a concrete step toward a world where deepfakes, synthetic media and chatbots must disclose what they are.
What kicks in August 2
The Act’s Article 50 transparency obligations arrive on a fixed date that has not moved: August 2, 2026. From then, AI systems that interact with people, generate synthetic content, or perform emotion recognition must comply with disclosure rules. In practice that means chatbots must identify themselves as AI, deepfakes and AI-generated images/audio/video must be marked, and emotion-recognition systems must tell users they are in use.
The new Code of Practice
To guide compliance, the European Commission published a Code of Practice on marking and labeling AI-generated content on June 10. It is a voluntary tool, but it signals how regulators expect providers and deployers to meet the transparency obligations — by clearly marking machine-made content and disclosing its artificial nature. Voluntary today, it sets the template for enforcement tomorrow.
Tougher measures coming
The rules sharpen further. A recent political deal added a ban on AI ‘nudification’ apps that generate non-consensual intimate imagery, and a mandatory watermarking obligation for AI-generated audio, image, video and text content, applicable from December 2, 2026. Rules for high-risk AI — in biometrics, critical infrastructure, employment, migration and more — apply from December 2027.
Why it matters
As generative AI makes synthetic media indistinguishable from real, the ability to know what is machine-made becomes essential to trust — in elections, journalism, commerce and personal life. The EU is betting that mandatory labeling and watermarking can preserve that trust. It is the most concrete attempt yet to make AI accountable for disclosing itself, and it will shape product design far beyond Europe.
The challenges
Enforcement is hard. Watermarks can be stripped, bad actors will ignore voluntary codes, and labeling does not stop a determined disinformation campaign. There are also free-expression and practical questions about marking text. The rules are a meaningful floor, not a cure — and their real test is whether they can be enforced against those who most want to evade them.
The bottom line
The EU AI Act’s transparency rules and new labeling code mark a turning point: from August, AI in Europe must increasingly say so. Deepfake labels, chatbot disclosure, watermarking and a nudification ban form the most serious regulatory push yet to keep synthetic media honest — a framework the rest of the world will watch, and likely borrow from.
Photo: MDGovpics / BY via flickr