The metadata tagging uses specifications from the Coalition for Content Provenance and Authenticity (C2PA), an organization that develops technical standards for certifying the source and history of media content. However, OpenAI admits that metadata like C2PA is not foolproof as it can easily be removed. The metadata is also not immediately visible to a casual observer, unlike Meta's AI labeling scheme, which will include a sparkles emoji as an immediate signifier that an image was made with AI tools.
Key takeaways:
- OpenAI is updating its ChatGPT and DALL-E 3 AI image generator model to include new metadata tagging, allowing users and organizations to identify images made with AI tools.
- The company has implemented the use of C2PA specifications for metadata, which will help in identifying AI-generated content.
- OpenAI's metadata is not immediately visible and can be easily removed, unlike Meta's proposed AI labeling scheme which will be public-facing and include a visual signifier.
- The company has also launched a website called Content Credentials where users can verify if an image is AI-generated or not.