The rise of generative AI has made deepfakes easier to create than ever. People like celebrities and content creators, who have countless hours of video footage of themselves online that can now be scraped and fed to software, are particularly at risk for bad actors. It may seem funny when gen AI is used to make videos of Joe Biden griefing opponents in Minecraft, but creators like QTCinderella have talked about the much darker side of widespread deepfake tech.
YouTube has been extremely bullish on AI, but at the same time as it’s introduced tools like Dream Track and automated chat summaries (which didn’t quite work out as expected), it’s promised that creators and their concerns are always on its mind.
In July, the platform took a step forward in combating deepfakes by saying it would remove AI-generated content that realistically simulated a person’s face or voice—if that person asked it to.
Subscribe for daily Tubefilter Top Stories
Now, its new partnership goes further: It’s teaming with Creative Artists Agency (CAA) to help build a detection system that will actively seek out deepfake content of celebrities and content creators so they can request removals.
It’s rare for YouTube to work with one of the many talent management companies and agencies that have sprung up around both Hollywood and the content creator industry—but in this case, it says working with CAA gives it access to “several of the world’s most influential figures.” It didn’t name names, but said the cohort includes “award-winning actors and top athletes from the NBA and NFL.”
All of those people have large amounts of footage of themselves available to the public, and thus are at high risk of being deepfaked.
“By collaborating with CAA, we’ll gain insight from some of the world’s most influential figures—some of whom have been significantly impacted by the latest waves of AI innovation—to refine our product before releasing it to a wider group of creators and artists,” YouTube said in a company blog post.
CEO Neal Mohan added, “At YouTube, we believe that a responsible approach to AI starts with strong partnerships. We’re excited to collaborate with CAA, an organization that shares our commitment to empowering artists and creators. In the days ahead, we’ll work with CAA to ensure artists and creators experience the incredible potential of AI while also maintaining creative control over their likeness. This partnership marks a significant step toward building that future.”
CAA’s participating clients will have access to “early-stage technology designed to identify and manage AI-generated content that features their likeness, including their face, on YouTube at scale,” the platform added.
Celebrities will provide feedback to YouTube while it’s building the tool, and then will be the first to use the live version when it rolls out in early 2025. YouTube says one goal here is to provide “easy access” to submit takedown requests when its tool does find deepfaked material. The takedowns will be categorized as privacy complaints.
We don’t think those takedowns will result in strikes for the uploaders, since YouTube is leaving it up to individual people to decide whether they allow deepfakes of themselves on the platform, instead of wholesale banning it.
This is in line with YouTube’s other policies around AI. It just introduced a “third-party toggle” for creators that will allow them to opt out of having their videos used to train generative AI models. Creators are opted out by default, but the toggle’s existence means YouTube will allow outside entities to scrape creators’ content, just like it will allow uploaders to post deepfakes.
YouTube says its partnership with CAA is “the first step of a larger testing effort,” and that over the next few months, it’ll announce new batches of testers, including “top YouTube creators, creator professionals, and other leading partners representing talent.”
