China Proposes Labels for AI-Generated Content
China’s Cyberspace Administration has put forward a draft regulation that would require all content created by AI to be clearly labeled. Announced on September 14, this proposal is expected to be enforced after a public comment period, which ends on October 14, 2024.
If approved, this regulation will help ensure that AI-generated content is easily identifiable, protecting the rights of citizens and organizations. The internet regulator explained that this would set clear standards for marking AI content.
The draft proposes two types of labels: explicit and implicit. Explicit labels are visual or audio cues, such as text or symbols, that make it obvious to users that they are viewing or hearing AI-generated content.
Implicit labels, on the other hand, would involve adding hidden data, like digital watermarks, to the content files. This would allow the source and nature of AI-generated content to be tracked.
Service providers will be responsible for applying these labels to all forms of AI media, whether it's text, images, audio, or video. Text-based content will need a clear warning at either the beginning or end. For audio files, there must be voice prompts to inform listeners. Images and videos will require visible indicators to show they were created by AI.
Additionally, implicit labels will have to include metadata with details about the content, such as the provider’s identity and a unique identifier for tracking purposes. Platforms that host AI-generated content will be responsible for verifying this metadata and ensuring that the appropriate labels are in place.
The proposal also includes penalties for those who attempt to remove or tamper with these labels.
China’s plan is similar to recent discussions in California, where lawmakers are also pushing for AI-generated content to include watermarks. This movement, supported by figures like Elon Musk, aims to make it easier to identify AI-created material through its embedded metadata.