Transparency requirements when working with AI

Do you use AI systems for your business? As a provider or deployer of certain AI systems, you have a transparency obligation. You must, for example, label your AI-generated content for your readers, users, or customers. 

When do I have a transparency obligation? 

Whether you have a transparency obligation depends on how you use the AI system. You must disclose that you are using an AI system if you use it to: 

• interact with users, such as a chatbot or telephone receptionist
• create or modify (manipulate) content, such as images, text, or audio
• recognise people's emotions or categorise them based their on biometric characteristics for medical or safety reasons
• create deepfakes, also when used for artistic or satirical purposes
• generate texts on matters of public interest without editorial (human) control 

If it is sufficiently clear to users of your system that it is an AI system, or a system that uses AI, you do not have transparency obligations. 

How do I comply with the transparency obligation? 

How you meet the transparency obligation depends on the type of AI-system you use: 

• A system that interacts with people, such as a chatbot, must actively state that it is an AI system.
• A system that creates artificial content must clearly mark the generated content, with a machine-readable watermark and label. 
• If you use an AI-system for emotion recognition or for biometric classification, you must inform the individuals whose data the system processes. 
• If you deploy a system to generate or edit deepfakes, you must visibly or audibly state that AI was used to create your content. 

Please note: from 2 December 2026, existing AI-systems must also mark their generated content with a machine readable watermark.