Art. 50 EU AI Act: Labeling Obligations for AI Content
From August 2026, AI-generated text, images, videos and audio must be labeled. Here's what you need to do.
What does Article 50 regulate?
Article 50 of the EU AI Act defines transparency obligations for certain AI systems. The aim is to inform people when they encounter AI-generated or -manipulated content.
The obligation applies to both providers (who supply the AI system) and deployers (who use it). Both must ensure AI content is correctly labeled – machine-readable and recognizable to users.
Labeling Requirements by Content Type
AI-Generated Text
Text created wholly or partly by AI systems – e.g. through ChatGPT, Claude or similar language models. This includes blog articles, product descriptions, social media posts and newsletters.
How to label correctly
Visible notice such as 'This text was created with AI assistance'. Additionally, machine-readable metadata (e.g. C2PA standard) recommended.
AI-Generated Images
Images generated wholly or partly by AI or significantly altered – e.g. through Midjourney, DALL-E or Stable Diffusion. Also includes heavy AI-based image editing.
How to label correctly
Label in the image itself (watermark) or as a caption. C2PA/Content Credentials as machine-readable standard. Metadata in EXIF header.
AI-Generated Videos & Deepfakes
Videos created or manipulated by AI, including deepfakes, synthetic speakers and AI-animated avatars. Particularly relevant for social media and advertising.
How to label correctly
Clearly visible labeling at the beginning and end of the video. For deepfakes: additional notice that persons are synthetically represented.
AI-Generated Audio
Audio content generated by AI – e.g. synthetic voices, AI-generated music, voice cloning or text-to-speech applications.
How to label correctly
Verbal or written notice that the voice/music is AI-generated. For podcasts/radio: labeling in the description and at the beginning.
Labeling Required
Chatbots and conversational AI: Users must know they are communicating with AI
Deepfakes and synthetic media: Always require labeling
AI-generated text that could be perceived as human-created
Emotion recognition and biometric categorization: Affected persons must be informed
Exemptions from Labeling
AI-assisted spell checking or grammar correction (pure auxiliary function)
AI content that has undergone editorial review and is published under human responsibility
AI use for law enforcement or national security (under certain conditions)
Practical Checklist: Implementing Art. 50
Create AI inventory
Document all AI systems in your company that create or modify content.
Define labeling standard
Choose a uniform standard (e.g. C2PA) and define templates for visible labeling.
Define processes
Determine who in the company is responsible for labeling and how the approval process works.
Train employees
Raise awareness among all content creators about labeling obligations and correct procedures.
Review regularly
Conduct regular audits to ensure all AI content is correctly labeled.
Penalties for Violations
Violations of transparency obligations under Art. 50 can be fined up to €15 million or 3% of global annual turnover (Art. 99(4) EU AI Act).
Source: Regulation (EU) 2024/1689, Article 50 – Transparency obligations for providers and deployers of certain AI systems. EUR-Lex Volltext
Check whether your AI content complies with labeling requirements.