The standard, though not completely reliable, means of blocking web crawlers on websites is the robots.txt file. Adobe intends to implement its equivalent for image files, so that their authors can set a ban on training artificial intelligence using these images.
Image source: blog.adobe.com
The solution is based on Content Credentials technology using the Coalition for Content Provenance and Authenticity (C2PA) standard to determine the authenticity of content. Adobe has released a web application that allows content authors to add a label to image files, even if these images have not been edited using the company’s programs. Here you can also set a mark that will signal AI companies to exclude such images from data sets for training models. The Adobe Content Authenticity App web application allows you to attach data from owners, including their names and links to social media pages, to sets of up to 50 JPG or PNG files at a time.
As part of the pilot program, Adobe has secured the support of professional social network LinkedIn, which has rolled out verification features — proof that the person attaching credentials has a verified LinkedIn page. Instagram✴ and X social network profiles can also be added to the metadata, but the platforms themselves do not provide integration with the new solution. The same web application allows you to set an option that will indicate that the image should not be used to train AI models. Although this field is present in the application, and the corresponding marking is added to the metadata, Adobe has not yet entered into an agreement with any of the AI developers to adopt the new standard, but negotiations are already underway with market leaders. The company has also released a Chrome extension that will make it easier to identify images with Content Credentials.
The Content Credentials web app uses several metadata schemes so that after editing, the metadata remains intact, and the Chrome extension will point out such an image even if it is published on Instagram✴, a platform that does not support the standard. If the Content Credentials fields in the image data are filled in, the user will see a small “CR” symbol on it. In the future, the company intends to implement similar markings in audio and video files.