▼ Summary
– Adobe is introducing a new tool within content credentials to give creators control over how their images are used in AI training, similar to robots.txt for websites.
– The tool, called Adobe Content Authenticity App, allows creators to embed metadata in their images to verify authenticity and ownership and indicate that their images should not be used for AI training.
– Adobe faces challenges in getting AI companies to comply with this new standard, as AI crawlers often ignore such requests.
– The app includes features like attaching credentials (e.g., name, social media profiles) to images and marking them as not intended for AI training, but Adobe has not yet secured agreements with AI model developers.
– Adobe is also launching a Chrome extension to help users identify images with content credentials and plans to extend support for videos and audio content in the future.
For many years, websites have utilized robots.txt files to specify which web crawlers are not permitted to access their pages. Adobe aims to establish a similar protocol for images, introducing a new tool within content credentials to give creators more control over how their images are used in AI training.
One of the biggest hurdles Adobe faces is persuading AI companies to comply with this new standard, especially given that AI crawlers frequently disregard requests in the robots.txt file.
Content credentials embed metadata in media files to verify authenticity and ownership. This is part of the Coalition for Content Provenance and Authenticity (C2PA), a standard focused on content authenticity.
Adobe is launching a web-based tool to allow creators to add content credentials to their image files, regardless of whether they were created or edited using Adobe tools. This feature also provides a means for creators to indicate that their images should not be used for training AI models.
The new tool, called Adobe Content Authenticity App, enables users to attach credentials—such as their name and social media profiles—to their images. Users can apply these credentials to up to 50 JPG or PNG files simultaneously.
Adobe is collaborating with LinkedIn to utilize its verification program, ensuring that the person attaching credentials to an image has a verified LinkedIn identity.
Additionally, users can link their Instagram or X profiles to an image, although there is no verification integration for these platforms.
The app includes an option for users to mark their images as not intended for model training. However, Adobe has not yet secured agreements with AI model developers to adopt this standard. The company is currently in discussions with leading AI model creators to encourage them to recognize and respect this indicator.
Despite Adobe’s positive intentions to provide a clear signal to AI model developers regarding training data, the success of this initiative depends on companies agreeing to and honoring the standard.
Last year, Meta’s introduction of labels to auto-tag images on its platform sparked controversy when photographers found their images labeled as “Made with AI.” Meta later revised the label to “AI info.”
This controversy underscored the differences in implementation of content authenticity standards among C2PA steering committee members, such as Meta and Adobe.
Andy Parson, Senior Director of the Content Authenticity Initiative at Adobe, explained that the new content credential app was developed with creators in mind. With copyright and AI training data regulations varying globally, Adobe aims to provide creators with a method to express their preferences regarding AI use of their content.
“Content creators want a straightforward way to indicate that they don’t want their content used for AI training. We’ve heard from small creators and agencies that they desire more control over their creations in the context of AI training,” Parson shared.
Adobe is also introducing a Chrome extension to help users identify images with content credentials.
The company stated that the app uses a combination of digital fingerprinting, open-source watermarking, and cryptographic metadata to embed information within an image’s pixels, ensuring the metadata remains even if the image is altered. Users can utilize the Chrome extension to view content credentials on platforms like Instagram that do not natively support the standard. Images with content credentials will display a small “CR” symbol.
Amid ongoing debates about AI and art, Parson emphasized that C2PA does not dictate what constitutes art. However, he believes content credentials can serve as significant markers of ownership.
“There is a grey area when an image is edited using AI but is not entirely AI-generated. We believe artists and creators should be able to sign their work and claim attribution. This doesn’t guarantee the IP is legitimate or copyrightable, but it indicates that someone created it,” Parson explained.
Adobe plans to extend support for videos and audio content with this tool in the future.
(Source: TechCrunch)
