With AI tech progress outpacing regulations to rein it in, digital artists have little guarantee that their work won’t be fed to a dataset that could ultimately help an image generator mimic their style.
Adobe is aiming to change that with a new tool that would give creators a means to digitally sign their images, videos, and audio files and potentially control how they are used.
The software giant’s new tool, set to roll out early next year, will allow creators to attach Content Credentials to their work, a “nutrition label” of sorts that includes attribution data and preferences around use in generative AI training.
The credential system is part of the Content Authenticity Initiative (CAI), an Adobe-led industry group that aims to create a standard for tagging the provenance of digital media at a time when sophisticated AI visuals have eroded a sense of reality online.
“We all read headlines about copyright lawsuits and GenAI models scraping the web with impunity, without any regard for ownership or compensation; these are real concerns that we’re hearing from our community,” Andy Parsons, senior director at the Content Authenticity Initiative, told Tech Brew. “Until there are copyright laws and court cases that protect that explicitly by law, we think it’s very important now and into the future beyond those times, to ensure that creators have agency to specify their preferences.”
Teamwork makes the scheme work: Like much of what the CAI does, some elements of the push involve cross-industry coordination. For one, AI companies need to agree to respect artist preferences. Parsons said Spawning AI, a data governance platform that counts companies like Stability AI and Hugging Face as partners, has already signed on to do so, and there are “many other conversations in late stages.”
“It’s not a hard negotiation in general, because we’re not negotiating a business contract here. We’re saying…the creator ecosystem desires this, and opt-outs here and there probably won’t impact the quality of a model that a commercial entity is building, and we think they should respect it,” Parsons said.
Keep up with the innovative tech transforming business
Tech Brew keeps business leaders up-to-date on the latest innovations, automation advances, policy shifts, and more, so they can make informed decisions about tech.
The CAI also plans to work with browser companies to eventually better support visibility for content credentials—Microsoft is among the CAI’s thousands of members, and Google has worked with the organization, Parsons said. For now, however, Adobe is releasing a browser extension that allows web surfers to see these tags on websites and social media platforms that don’t already display them.
Digital durability: But couldn’t would-be content thieves simply remove the content credential tag or screenshot an image? Parsons said it’s not so easy with the tech that Adobe has designed to make this attribution data sticky. The company uses a combination of “the best parts of watermarking, something called fingerprinting, and secure metadata” to ensure that credentials can’t easily be tampered with.
“It marries them together in a way that makes this data survive screenshotting or taking a picture of a picture, which is a way that people often try to strip metadata, or try to pass off a photo of a screen as a real photo,” Parsons said.
Originally founded in 2019, the CAI and its linked standards org, the Coalition for Content Provenance and Authority (C2PA), have been attracting membership as the threat of sophisticated deepfakes has gotten more real. Membership of the two orgs spans tech giants like Microsoft, Google, and Adobe; media outlets such as BBC, the New York Times, and the Associated Press; and camera companies including Leica, Canon, and Nikon.
Update 10/13/24: This story has been updated to clarify the nature of Hugging Face and Stability AI’s involvement.