A Helpful Decision Tree for AI Labelling under Article 50 of the EU AI Act

I have recently been looking more closely at when AI-generated content actually needs to be labelled. Especially when it comes to text and images, the answer is far more nuanced than it may initially seem. In this context, I came across this decision tree created by Markus Begerow. It provides a clear overview of the key questions that companies, public authorities, editorial teams and content managers should consider before publishing AI-generated content. What I find particularly helpful is the clear distinction between text and images. For AI-generated text, the relevant questions include whether the content concerns a matter of public interest and whether it has subsequently been reviewed by a person or placed under editorial responsibility. For images, the main issue is whether real people, places, objects or events are depicted or manipulated in a deceptively realistic way. Where the content qualifies as a deepfake, visible labelling may be required. https://preview.redd.it/sqjfc2uq87hh1.png?width=2400&format=png&auto=webp&s=40f4a96d661e4a0d67b5d0df3181186bc1c874a0 From my perspective, the graphic highlights one important point: not every piece of AI-generated content automatically requires visible labelling. The decisive factors are the specific content, the context in which it is published and its potential to mislead. For me, this decision tree is therefore a very useful first point of reference for understanding Article 50 of the EU AI Act in a practical and accessible way. Thank you to Markus Begerow for presenting this complex topic so clearly. How are companies, public authorities and editorial teams currently approaching AI labelling in practice?