OpenAI incorporates watermarks into the descriptive data of AI-generated images produced by DALL-E 3, as more companies join in supporting the C2PA alliance standards.
The company explains that the watermarks from C2PA are present in the images generated through the ChatGPT website and the DALL-E 3 model’s API.
Mobile phone users receive watermarks on February 12th. These watermarks include invisible descriptive data and a visible Content Credentials icon, which appears in the top left corner of each image.
Individuals can verify the source – any artificial intelligence tool used to generate content – of any image generated by OpenAI platforms through websites such as Content Credentials Verify.
Currently, artificial intelligence companies have the ability to add watermarks to still images, but not to videos or texts.
OpenAI states that adding descriptive metadata to the watermark of images has minimal impact on access time and will not affect the quality of image generation. Additionally, the addition leads to a slight increase in the sizes of the images for certain tasks.
The C2PA alliance encourages the use of Content Credentials metadata to identify the source of content and determine whether it was created by humans or artificial intelligence.
Adobe created the Content Credentials icon, and OpenAI plans to incorporate this icon into the images generated using DALL-E 3.
Meta also announced that it is adding tags to AI-generated content across its social media platforms.
The artificial intelligence content moderation is considered one of the main priorities in Biden administration’s executive order on artificial intelligence, although watermarking is not a guaranteed method to prevent false information.
OpenAI states that the descriptive metadata of the C2PA alliance standards can easily be removed accidentally or intentionally, especially since most social media platforms often remove descriptive metadata from uploaded content. Taking a screenshot results in the deletion of the descriptive metadata.
OpenAI states on its website that they believe adopting these methods to identify the source and encourage users to recognize these signals is the key to increasing the credibility of digital information.
