Despite the new measures, OpenAI acknowledges that the plans are not perfect. The company plans to extend the inclusion of C2PA in the metadata for images produced with the web version of DALL·E 3 to mobile users by February 12. However, it notes that the metadata is not a definitive solution to address issues of provenance as it can also be easily removed. The company has previously taken down its detection service due to accuracy concerns.
Key takeaways:
- OpenAI is adding new digital watermarks from the Coalition for Content Provenance and Authenticity (C2PA) to its AI-generated images, including those from DALL-E 3, in an effort to increase public trust in digital information.
- Users can verify if an image has been AI-generated through sites like Content Credentials Verify, and some media organizations are using the standard to verify content sources.
- Despite these measures, OpenAI acknowledges that the method is not foolproof as the watermark could be easily removed, and there is increasing concern about the spread of AI-generated misinformation, especially in the context of elections.
- Meta, another tech company, is also planning to tackle the spread of AI-generated content by attaching labels to such images on Facebook, Instagram, and Threads.