What Should Rights Holders Know About Generative AI?
Jun 5, 2023
Who This Is For
This primer is for (a) anyone at studios or labels in in the film, television, and music industries, (b) talent including actors, singers, and writers and (c) rights holders and estates. Generative AI will impact your creative works and intellectual property, including AI-generated text, images, videos, and audio — have you heard of the new AI-generated Drake & The Weeknd song or read how Grimes is allowing anyone to use her voice to create AI-generated music?
Overview of Key Legal Concepts
Copyright law grants creators the exclusive right to control the use and distribution of their work. Unlike trademarks and patents, copyright automatically protects original works of authorship fixed in a tangible medium (such as music, literature, films, and software) and does not require registration; however, there are certain benefits to registration such as (i) the ability to bring an infringement action, (ii) to evidence the validity of the creator’s copyright, and (iii) to create a public record so others are put on notice that the creator is the copyright owner. As a general rule, for works created after January 1, 1978, copyright protection lasts for the life of the creators plus an additional 70 years, though this does vary by jurisdiction. Infringement of copyright can lead to legal action, including monetary damages, injunctions, or criminal penalties.
Copyright law is rooted in the desire to incentivize creative works by protecting the creator’s economic rights, and was developed after the invention of the printing press.
Trademark law protects distinctive signs, symbols, or names used to identify and distinguish a well-known or famous person, someone with public recognition in the relevant industry, business entity, goods, or services. Unlike copyright law, which automatically protects creative works upon creation, trademarks must be registered with the United States Patent and Trademark Office (USPTO) to receive protection. The USPTO allows registration of names, signatures, and portraits that could identify a particular living individual, but not their voice. Trademark infringement can result in legal action, including monetary damages, injunctions, or cancellation of the infringing mark's registration.
Trademark law is rooted in the desire to eliminate customer confusion about the source of a product and to protect brands from unfair competition.
Likeness (Publicity Rights)
Publicity rights are an individual's exclusive right to license the use of their name, likeness, signature, voice, or other recognizable aspects of their persona for commercial benefit. For instance, back in 1988, the Ninth Circuit held that Ford Motors had infringed on Bette Midler’s right of publicity when Ford used a Bette Midler-sound alike singer to sell cars in one of their commercials.
In the US, the right of publicity is largely protected by state common or statutory law (no federal statute recognizes this right). Over half of the states recognize the right of publicity, and many of those do not recognize a right by that name but protect it as part of the right of privacy, which protects a person from the publication of embarrassing private facts, from being put in a false light, and from publication of false information. When it comes to student-athletes in particular, half of the states have passed “Name, Image and Likeness (NIL)” laws that protects the rights of student-athletes to commercialize their name, image, and likeness.
Contracts play a significant role in the protection and commercialization of intellectual property rights by outlining the responsibilities and obligations of the contracting parties involved in the creation, distribution, and protection of creative works. For instance, IP license agreements allow one party to utilize the other party’s IP property (including patents, trademarks, copyrights, and trade secrets). Personal or talent release forms permit the use of a person's name, image, or likeness, and can be in the form of a (i) blanket release, which gives an unlimited right to use their likeness, or a (ii) limited release, which specifies particular ways their likeness may be used.
Fair Use and the First Amendment
Fair use is a defense under the First Amendment that is typically raised in response to a claim of copyright infringement. In other words, it’s a legal concept that permits the use of copyrighted material without permission from the copyright owner. Determining whether a use of a copyrighted work can be complex, and courts will consider factors such as: (i) the purpose and character of the use (i.e., how “transformative” is the new work from the original work), (ii) the nature of the copyrighted work (i.e., factual or creative, published or unpublished), (iii) the amount and substantiality of the portion used (i.e., small portion of the work vs. the “heart” of the work, but parody is generally excused), and (iv) the effect of the use on the potential market for the copyrighted work.
Some, if not all, of the elements of the fair use doctrine are also considered in determining whether a particular use of a person’s likeness is fair. Transformative use is a defense to using someone’s likeness in a way that creates a new meaning or message, such as using a celebrity’s image in a work of art that is commenting on celebrity culture as a parody and/or to provide cultural and entertainment value. The First Amendment can also provide a defense if the use of a person's identity is in the public interest (also known as a use’s “newsworthiness”) or contributes to public debate. For example, the Ninth Circuit in 1999 held in Hoffman v. Capital Cities/ABC, Inc. that Los Angeles Magazine’s publication of an article containing a digitally altered photograph of actor Dustin Hoffman so it appeared he was wearing certain designer clothing was protected by the First Amendment because “the photograph consisted of a combination of fashion photography, humor, and visual and verbal editorial comment on classic files and famous actors” and thus was in the purview of public interest. The court further concluded that the magazine’s alteration of Hoffman’s image did not constitute a knowing falsehood because readers would know that the picture was digitally altered.
Is it legal to use AI to generate someone’s voice or image?
Does this infringe upon the person’s right of publicity?
The legality of using AI to generate someone's voice or image under the right of publicity depends on several factors, including the individual's domicile, the identifiability of the AI-generated voice or image, and the purpose of the use. In order to establish a cause of action for a right of publicity violation, a plaintiff must show: (1) the validity of the plaintiff’s right of publicity, and (2) that this right has been infringed upon by the defendant.
Validity of plaintiff’s right of publicity
Under this first prong, the first step is determining whether the person is (or was at the time of death) domiciled in a jurisdiction that provides right of publicity protection. For instance, NY passed new legislation in the past 2 years that provides protection after death of the individual for 40 years, whereas under CA legislation the length of time is 70 years. That means that within that time frame, prior consent is required to use the deceased personality’s name, voice, signature, photograph or likeness in association with products, merchandise or goods (or for purposes of advertising or selling these).
Next is determining whether the relevant jurisdiction places any restrictions on which individuals qualify for right of publicity protection; for instance, some state statutes specify that the right extends only to public figures, while others provide a right of publicity only for individuals who have served in the armed services.
Finally, in the case of a deceased individual, because most jurisdictions recognizing a postmortem right of publicity extend the right only for a fixed period following the death of the individual, one must consider the date of death in order to determine whether the right is still protected in the relevant jurisdiction.
Right of publicity has been infringed upon
Under this second prong, courts generally take the following approach that consists of the following 2 elements: (1) the defendant, without permission, has used some aspect of the plaintiff’s identity or persona in such a way that the plaintiff is identifiable from the defendant’s use, and (2) the defendant’s use is likely to cause damage to the commercial value of that persona.
The first element is also known as the test of identifiability. Applying this to an AI-generative work, the inquiry would be whether an AI-generated voice or image can be linked to (i.e., is sufficiently similar to) a particular individual, even if the voice or image is not explicitly identified as theirs. For example, if an AI-generated voice is identifiable as Taylor Swift in the minds of the public, then the use of that voice may be considered a violation of her right of publicity, even if the voice is not explicitly labeled as being hers.
If under the first element the voice or image can be linked to a specific individual, then we move on to the second element of commercial vs. non-commercial use. A use that commercial, such as to endorse a product or service (like in the Bette Midler case), would be a near-certain violation of this person’s publicity right.
Non-commercial uses could also give rise to claims of defamation or false light if the AI-generated voice or image is used to defame or misrepresent an individual. Deepfakes, which are videos in which a person's face or body has been digitally altered so that they appear to be someone else, raise additional concerns related to malicious use, false information, and revenge porn laws.
First Amendment Defense
The “transformative use” or “newsworthiness” defense under the First Amendment to a right of publicity violation claim of an AI-generative work inquires: (1) was the AI-generated voice or image used in a way that creates a new meaning or message, such as in a work of art or political commentary, and (2) was the AI-generated voice or image used in a way that is newsworthy or in the public interest, such as in reporting on a current event or historical figure? If the answer to either of these is “yes,” then the AI-generative work would not be found in violation of the individual’s publicity right.
Does this infringe upon the copyright of source material used?
The legality of AI-generative work (let’s call this the AI Work) using someone else’s copyright (let’s call this the Prior Work) without their consent depends on two main factors: (1) whether the allegedly infringing work took elements from the original work, and (2) whether there is substantial similarity between the original work and the allegedly infringing work.
Prior Work elements
Under this first prong, if the voice or image used in the AI Work was initially copied from the Prior Work to be altered by AI, the copying alone is technically copyright infringement, even if the AI Work does not resemble any elements of the Prior Work. However, most courts would likely apply the fair use defense to permit the initial copying.
Under this second prong, if the AI Work does substantially resemble elements of the Prior Work such that, for instance, the AI Work uses a voice or image from the Prior Work that he public associates with that particular person, the owner of the Prior Work may have a valid claim for copyright infringement. However, if the AI Work is sufficiently different from the Prior Work and does not incorporate any protectable elements of the Prior Work, it may not.
Fair Use Defense
Under the fair use factors described above, courts may deem the use of the copyrighted material in the AI Work without the owner’s consent as fair if it promotes creativity, education, and free speech.
a. The purpose and character of the use (transformative use)
Commercial use is generally weighed against the finding of fair use, while transformative use favors it. Transformative use occurs when the result of the AI Work and the Prior Work serve different market functions or offer something new and different from the original work.
Regards to AI, training a large learning model (a “LLM”) on copyrighted data is likely transformative for fair use purposes if it has a different purpose than the original work or constitutes copying for analysis or reverse engineering (discussed more below). However, if the AI Work serves the same purpose as the training data (aka the Prior Work), it is likely not transformative.
b. The nature of the copyrighted work
Under this factor, courts will look at whether the Prior Work is creative or factual in nature, and whether the Prior Work is published or unpublished. Factual works are often afforded less protection under copyright law, and fair use is more likely to be found in cases where the use is for educational or informational purposes. On the other hand, creative works are given greater protection, and fair use may be harder to establish when the use involves commercial exploitation of the work. Fair use is also more likely to be found with published works, as opposed to unpublished works, which are given stronger protection.
c. The amount and substantiality of the portion used
An important factor in determining amount and substantiality of the portion used is whether a lot was copied from the Prior Work and whether much of the AI Work consists only of the copied material. For example, fair use is unlikely if the AI Work used so much of the Prior Work that the AI Work is now a “competing substitute” that is available to the public.
d. The effect of the use on the potential market for the copyrighted work
The primary inquiry here is this: does the use compete with the Prior Work and its derivative works for its market? If yes (for instance, the use does harm the copyright owner’s ability to sell or license their work), then fair use is unlikely. However, this analysis also takes into account the benefit gained by the public if the use is held to be fair. If, for example, AI is used to generate a piece of music that is so similar to the Prior Work that it serves as a substitute for that work, then the AI Work would likely harm the copyright owner's ability to sell or license their work. On the other hand, if the AI's output is sufficiently transformative and does not compete with the Prior Work or its derivative works, then the use may not harm the copyright owner's ability to sell or license their work. For example, if AI is used to generate a new piece of music that is based on the Prior Work but is significantly different from the original, then it may not harm the market for the Prior Work and may be considered fair use.
Does training AI models with copyright works constitute infringement?
This is the question that is currently being contemplated in the 2023 class action lawsuit Andersen v. Stability AI et al. In this case, three artists sued a number of AI companies (Stability AI, Midjourney, DeviantArt) that used copyrighted images to train models for AI image generation products without consent from or compensation to the artists, who are the underlying image right-holders. Applying the elements of copyright law and the fair use defense, the court will be looking at the following questions:
How transformative is the AI generated work?
How much, quantitatively or qualitatively, was taken from any underlying works?
How does defendants’ distribution of the new AI generated work impact the market for plaintiffs’ works?
The plaintiffs, however, have not alleged that the defendants produce images which depict any individuals or objects identical to those found in plaintiffs’ photos; in fact, they seem to acknowledge that the defendants do not replicate elements of plaintiffs’ work in this sense. Furthermore, if the use serves a different purpose than the original works such that it cannot be reasonably seen as a “competing substitute” made available to the public, the court may find that the defendants’ use of the plaintiffs’ photos is fair.
Can one copyright AI-generated output?
Currently, the law does not offer copyright protection for works generated solely by a machine. The US Copyright Office states that any work must be human-made to be copyrightable. “Human” means both humans and corporate beings, and if the creator can prove there was a substantial human input, then the creator will own the IP. In other words, a computer program or AI itself can never be the author or owner of IP.
What rights would need to be negotiated if I were to allow someone else to use my likeness?
For someone else to use your likeness, you will need to negotiate the right to use your likeness (name, image, voice, and so on) or life story, depending on the context (for example, if you were acting in a movie or TV show). If someone wants to use your voice from a particular source (like a song), you, or the record label that owns the masters, will need to negotiate the copyright to the sound recording. If your name or brand is associated with registered trademarks, you may also need to negotiate the right for someone else to use those trademarks. Additionally, the party using your likeness may negotiate releases to protect themselves from defamation and invasion of privacy suits.
Any licensing of these rights would also include the negotiating of: (i) the scope and duration of the license to use your likeness; (ii) specific types of AI generative works in which your likeness can be used and/or the exclusion certain types of uses altogether; and (iii) compensation (e.g., a royalty or flat fee), among other protective provisions such as indemnification, termination, confidentiality, and dispute resolution.
Key Concerns & Opportunities
With the rise of AI generative works, artists should be concerned about the unauthorized use of their voice and likeness from fans or unauthorized commercial use by studios or labels in these works. It is important for artists to protect their rights and negotiate any necessary agreements for the use of their likeness. Contracts should be reviewed carefully to ensure that they do not contain clauses that give away full publicity or simulation rights (which are often times buried deep in boilerplate talent agreements), particularly for newer talent with lesser bargaining power.
On the other hand, artists should also look for opportunities to commercialize their likeness for AI commercial use by licensing them out in a limited fashion. This can provide a new revenue stream for artists and help them retain control over their intellectual property.
Another approach is to "open source" their voice and likeness, which is what Grimes is doing when she announced that anyone can use her voice for AI-generative music, and she’ll split profits 50/50 if any of the songs become “successful” through her new AI voice software. This approach may allow artists to make their voice and likeness available to the public for creative use, while still maintaining some control over how they are used. Keep in mind that the software is currently in beta, so stay tuned for further updates on the viability of this approach.
Finally, artists should stay informed about the latest developments in AI generative works and be proactive in protecting their intellectual property rights. They should consult with legal professionals to ensure that their rights are protected and negotiate favorable terms in any contracts related to the use of their voice and likeness in AI generative works.
For studios, production companies, and labels
Studios, production companies, and labels also face concerns when it comes to AI generative works. One key concern is the potential for legal challenges from copyright holders or the estates of deceased artists or performers. In such cases, licensing agreements must be carefully negotiated to ensure that all necessary rights have been obtained.
Other legal and ethical concerns include: (i) the potential for generative AI to create works that are similar to existing copyrighted works, which could lead to accusations of copyright infringement, and (ii) privacy issues. They must be careful to ensure that any works created using generative AI are sufficiently original and transformative to avoid such claims for the former, and obtain explicit consent from talent to use their likeness in AI generative work through a licensing agreement or release that clearly explains the purpose of the work and how talent’s IP will be used.
The opportunities here, though, include (i) creating content in partnership with the estates of actors and artists while keeping in mind what licenses they need to obtain (e.g., likeness, life story rights, trademark rights, etc) and (ii) negotiating talent’s simulation rights to use generative AI to “patch” films, music, and ads. This means that they can then generate new content from existing works, giving them the ability to create new content more quickly and cost-effectively.
As AI technology continues to evolve, there may be a need for new regulations or standards to ensure that personal data and intellectual property is being handled appropriately. Studios, production companies, and labels should consult with legal professionals and keep an eye on any emerging regulations in this area and ensure that they are complying with all applicable laws and guidelines.