EU Mandates Deepfake Labels Under the AI Act

Starting this Sunday, AI companies that provide generative systems such as Anthropic’s Claude and OpenAI’s ChatGPT must ensure that deepfakes and other AI-generated content are clearly identifiable as artificial. The rules apply to providers and to organisations that use AI in a professional capacity; personal, artistic, satirical or fictional creations are excluded. The requirement covers realistic AI-generated images, audio or video that could be mistaken for a real person’s actions or statements.

Euronews

The European Union adopted the AI Act two years ago and created a dedicated AI Office to oversee its implementation. Among the Act’s provisions is the world’s first transparency regime for synthetic media, obligating providers to make AI-generated or manipulated content detectable. To operationalise the regime, the European Commission drafted a voluntary code of practice that outlines how companies should disclose the artificial origin of synthetic content through machine-readable markings and visible labels for deepfakes.

Euronews

Three major hurdles affect the rule’s effectiveness. First, the EU can only regulate firms operating in or targeting the European market, while deepfakes produced elsewhere can reach European users instantly. Second, major AI firms—including OpenAI and Google—have signed the code but warn that detection and marking technologies remain a moving target; no single industry-wide solution exists, and companies are experimenting with disparate approaches such as watermarking, metadata and provenance systems that do not interoperate. Third, digital traces are fragile: watermarks can be removed, altered or lost during editing, compression or cross-platform sharing, making it difficult to trace the full history of a synthetic image or video.

Euronews

The EU’s aim is to give citizens the ability to recognise AI-generated content, thereby reducing the risk of political manipulation (e.g., a fake 2022 video of Ukrainian President Volodymyr Zelenskyy) and non-consensual explicit imagery (such as the 2024 AI-generated images of pop star Taylor Swift). The European Commission’s technical study concludes that a multilayered approach—combining several marking and detection methods—offers the strongest protection. However, analysts note that the rapid evolution of AI technology will test whether the transparency rules become an effective safeguard or merely a label that struggles to keep pace.

Euronews