EU’s Transparency Code of Practice: Watermark To Warning Label

Article 50 is applicable to both providers and deployers of certain AI systems. However, the exact obligations differ. Providers must ensure all AI-generated or manipulated content is appropriately marked and detectable. Deployers must disclose the artificial origin of some content produced by their models, such as deepfakes and AI-generated text, on matters of public interest.

AI content marking and AI content labeling serve different transparency purposes. Marking is the provider’s responsibility and requires embedding machine-readable transparency measures into AI-generated content. Labeling is the deployers’ responsibility and involves providing a clear, easy-to-understand disclosure that informs individuals when content has been artificially generated or manipulated.

Yes, there are several exemptions based on the type of content and context in which AI is used. For example, disclosure obligations for deepfakes may be limited when the content forms part of a creative and artistic work. Similarly, some transparency requirements do not apply when AI systems are used for law enforcement activities.