Underscoring the growing debate about ownership of ones likeness in the age of AI and Deepfakes, Tom Graham, CEO of generative AI tech firm Metaphysic — the company known for popularizing a deepfake TikTok account spoofing Tom Cruise — has submitted for copyright registration his AI likeness with the U.S. Copyright Office. He seeks to demonstrate how copyright laws may be used to protect against the unpermitted creation and dissemination of content in which a person’s likeness is duplicated through AI.
As Hollywood grapples with ethics and legal questions surrounding use of the tech, the U.S. Copyright Office affirmed in March that most works generated by AI are not copyrightable but clarified they qualify for protection in certain instances.
Speaking on a panel in January during CES, national executive director and chief negotiator of SAG-AFTRA Duncan Crabtree-Ireland emphasized the importance of addressing this issue. He reported that the Guild is working on policies aimed at how to create opportunities — not take them away — while protecting its members from having their likenesses exploited.
In the United States, copyright laws don’t protect works solely created by machines. Various courts have ruled that copyrights can only be granted to works created by humans. An application containing AI-generated material can support a claim only if it satisfies a requirement that a human contributed to most of the work, like if a person “selected or arranged” the piece in a “sufficiently creative way that the resulting work constitutes an original work of authorship.”
In his copyright application, Graham lists himself as the author of the work. This is a crucial distinction from a copyright application by Stephen Thaler, the CEO of neural network firm Imagine Engines who listed an AI system as the creator of an artwork called A Recent Entrance to Paradise. The copyright office denied the registration, finding the work “lacks the human authorship necessary to support a copyright claim” and that “the nexus between the human mind and creative expression” is a crucial element of protection. Last year, Thaler sued the office, challenging the rejection.
Unlike Thaler’s application and others seeking protection for AI-assisted works, Graham stresses his contribution in creating the video for which he seeks a copyright. He says that he curated the dataset used to train the model and that he “added other manmade work” to successfully get the Metaphysic AI tool to produce a realistic image of himself. There was also editing and compositing work in the process, he notes.
“All of these steps are very deliberate, human-driven activities, which I believe are indistinguishable from anyone using Photoshop or VFX tools,” Graham says. “We’re using AI, but there’s a large degree of manmade work and human effort. It’s fundamentally me from top to bottom.” Perhaps most recognized for powering the @DeepTomCruise account, Metaphysic develops generative AI tools and services for talent, including through a partnership with CAA.
Viability of a copyright will depend largely on how the AI tool operates and how it was used to create the final work. If an AI tech receives solely a prompt and produces a work, for example, it’s not eligible for protection because the traditional elements of human authorship are determined and executed by the technology.
Standards for copyright protection is still up in the air. Intellectual property lawyer Suzanne Hengl, who does not represent Metaphysic, is skeptical Graham’s application will satisfy the office’s current standards for human authorship. “To the extent that there was extensive training for the AI tool to work and create an output, that won’t get you over the hump,” she says.
The partner at law firm Baker Botts notes that there may be protection if there was work completed by a human after the video was produced. She points to the copyright office’s guidance that an artist who uses Photoshop to edit an image qualifies for protection of the modified image.
While Graham’s video may not be copyrightable, the chief executive’s application highlights the lack of protection against the illicit dissemination of videos and photos in which a person’s likeness is duplicated through AI. There are no copyright laws specifically designed to combat the use of deepfakes. In fact, they permit them in some cases.
Several instances of deepfakes in Hollywood fall under the “fair use” exception to copyright infringement. The doctrine allows for unlicensed use of copyrighted works under certain circumstances, like commentary, criticism and news reporting. One of the four factors in whether a work qualifies is the purpose and character of the use, which extends protection for transformative works when the purpose of the copyrighted work is altered to create something with a new meaning or message.
The fair use exception to copyright law is why Kendrick Lamar was on legally firm ground when he used deepfakes to transform into Will Smith, Jussie Smollett and Kobe Bryant in “The Heart Part 5” music video. They didn’t consent to being in the video but the deepfakes would likely be considered transformative use to explore Black representation in Hollywood, among the music video’s other themes. They also could’ve been categorized as parodies, which is also protected as transformative use under copyright laws.
Among the reasons that Graham says he moved for registration of the AI-created video of himself is to forge a path for people to protect their identities, performances and brands against unpermitted content created using AI tools, like deepfake pornography. Under the Digital Millennium Copyright Act, a copyright holder can quickly take down material that infringes on their copyrights.
“I want to underscore the value of using the DMCA as a means to take down an infringing video within 24 hours,” says Jordan Manekin, a lawyer for Metaphysic. “That’s a much more readily available remedy for most people than just going to court and taking several years litigating.”
But even if protection is granted, Hengl says it’s unlikely Graham can use his copyright to take down AI-created photos and videos himself. “In terms of being able to use copyright as a way to enforce rights with respect to a likeness or to enforce privacy rights, it’s really trying to fit a square peg into a round hole,” she suggests. “You can only enforce protection against works that are substantially similar. There needs to be more than that person’s image — their attire, expression on their face, the pose their body is in, the background they’re in — that’s similar.”
Hengl notes that the traditional avenue of action against deepfakes and similar tech has been pursuing claims for violations of rights of publicity and privacy.
As SAG-AFTRA’s Crabetree-Ireland summed up at CES, “As this technology democratizes and becomes more accessible and available, this is going to turn into a broader public policy issue.”
