Sebastian Schrittwieser was interviewed by Studio 2 to explain the AI Act's new transparency obligations. The key points from the interview:
Chatbots must identify themselves. Anyone interacting with an AI system in writing or by voice must be told, visibly or audibly, at the point of first contact. A note in the terms and conditions is not enough.
AI-generated images require a label. Providers must mark their outputs technically, for instance through metadata. A label perceptible to humans is required wherever companies distribute photorealistic AI content. The example discussed in the interview was a property listing with an AI-furnished living room.
Private use is exempt. The labelling duties apply to professional use. Someone who generates a joke picture of a family member and shares it in the family WhatsApp group is not covered and does not have to add a label. The exemption is read narrowly, though: it ends where content is aimed at a wider public or serves a professional or political purpose. And it says nothing about whether the image itself is permissible. The right to one's own image, protection of reputation and data protection law continue to apply.
In Austria, the first point of contact for questions is the AI Service Desk at RTR.
Watch the full interview (German): https://on.orf.at/video/14333611/16131016/talk-neue-eu-verordnung-ai-act-fuer-mehr-ki-transparenz