Watermarking for AI content: What changes with new EU regulations

New European Union regulations mandate the labeling of many AI-generated contents, taking effect on August 2. These rules aim to increase transparency and prevent manipulation by clearly distinguishing between human and artificial intelligence interactions.

Under the new guidelines, users must be informed when they are communicating with an AI, such as a chatbot, rather than a human. This applies particularly to content created for professional purposes, including advertising and influencer marketing, where creators earn money from their content. Influencers who use AI in their work must adhere to these labeling requirements.

The regulations specifically target "deepfakes" – highly realistic manipulated content that can be used for deceptive purposes. For deepfake videos, labels should appear at the beginning, remain visible, or be re-displayed periodically. Interactive elements, like a pop-up when hovering over a manipulated face with a mouse, are also recommended for manipulated images.

For audio content, such as AI-generated podcasts, a spoken notice at the beginning is required. Text-based AI content should feature labels either above the text or near the headline. Additionally, the creation process for AI content should embed machine-readable digital watermarks or fingerprints, allowing other systems to reliably identify it as artificially generated.

These rules do not extend to AI-generated images posted on private social media profiles. However, AI providers themselves, such as OpenAI, Google, and Meta, are responsible for implementing these labeling mechanisms. The European AI Office is also scrutinizing large AI models for adequate control and security measures, requiring immediate reporting of any incidents, especially concerning potential cyberattacks initiated by the AI itself.