— Vejay Lalla, Adine Mitrani and Zach Harned
Advances in “deepfake” media techniques that use deep learning AI—from uncanny impersonation videos of Tom Cruise and other celebrities to immersive technologies that make it possible for gamers to enhance the avatars they create—are disrupting the entertainment industry. We look at some of the pros and cons of these new capabilities and point out relevant legal issues for both entertainment clients and their lawyers to consider.
Last fall, the U.S. Copyright Office issued a final rule adopting exemptions to the Digital Millennium Copyright Act provision prohibiting circumvention of technological measures controlling access to copyrighted works. We examine notable new or expanded exemptions, including one involving free and open source software (FOSS) that should be of interest to technology-based companies.
Expungement and Reexamination and Shorter Trademark Office Action Response Periods, Oh My! — Crystal Prais
What Does “Prior Art” Mean in Copyright Law? — Diana Buck
Director of the USPTO Grants First Director Review of an IPR to Samsung for Lithium-ion Battery Patent — Howard Lithaw Lim
Beware of Different Limitations Periods in Employee Trade Secrets Disputes! — Stuart P. Meyer
By Vejay Lalla, Adine Mitrani and Zach Harned
Ever since the first Terminator movie was released, we have seen portrayals of robots taking over the world. Now we are at the beginning of a process by which technology—specifically, artificial intelligence—will enable the disruption of the entertainment and media industries themselves.
From traditional entertainment to gaming, this article explores how deepfake technology has become increasingly convincing and accessible to the public, and how much of an impact the harnessing of that technology will have on the entertainment and media ecosystem.
What is a “Deepfake” and Why Does it Matter?
The term “deepfake” refers to an AI-based technique that synthesizes media. This includes superimposing human features on another person’s body—and/or manipulating sounds—to generate a realistic human experience. Actor Val Kilmer lost his distinctive voice to throat cancer in 2015, but Sonantic’s deepfake technology was recently used to allow Kilmer to “speak.” (The actor’s son was brought to tears upon hearing his father’s “voice” again.)
Deepfakes have also been used to break down linguistic barriers, including by English soccer great David Beckham in his Malaria No More campaign. There, deepfakes enabled Beckham to deliver his message in nine different languages. And sometimes deepfakes are used for downright fun, such as in this art installation, which allows users to take a “surreal” selfie with Salvador Dalí.
Leveraging Deepfakes to Enhance a Talent’s Skillset
Commercial applications of deepfakes currently include both hiring the underlying “deepfake actors,” as well as individuals whose likeness is used as a “wrapper” (i.e., the visage or likeness portrayed in the content) for the underlying performance. Where the so-called wrapper is a famous personality, this may save the underlying talent hours of time that would otherwise need to be spent on set; that burden can be shifted to the deepfake actor instead. Additionally, such technology allows influencers to create personalized messages for hundreds or thousands of individuals without the need to actually record each message.
The foregoing novel applications of this technology do not fundamentally change the nature of talent agreements or acquiring the necessary rights from talent—however, they do introduce new wrinkles that both negotiating parties must consider carefully. For example, control over the use of the talent’s likeness rights is always negotiated in great detail, but it is unlikely that talent releases or agreements generally contemplate the right to use likeness rights as a wrapper to generate a potentially infinite number of lifelike deepfakes. Additionally, morals clauses will require careful drafting to address whether a deepfake performance, potentially one in which the talent had no control, can serve as grounds to trigger termination. Talent unions may also have to consider more specifically how this technology is addressed in future industry negotiations.
Finally, there is the open question of whether this technology will help or hurt talent overall. On the positive side, the scalability of allowing an actor to appear in commercials or on websites for e-commerce all over the world (without requiring trips to the studio, learning a new language or improving accent work) could be empowering. For instance, Synthesia recently did this with two commercials featuring rapper and entrepreneur Snoop Dogg. The initial commercial was such a success that the company’s subsidiary wanted to use the same commercial, but with the branding and names switched out. Rather than having to reshoot, Synthesia used deepfake technology to change Snoop Dogg’s mouth movements to match the subsidiary’s name in the new commercial.
On the other hand, the widespread adoption of deepfakes could allow for the supplanting of actors who are not celebrities, leading to job losses or a shift in how the industry hires talent for productions. If it becomes more efficient and otherwise desirable to hire relative unknowns to portray those with celebrity status, there are fewer opportunities for these actors to become known or “get discovered” in their own right. That could lead to the creation of a caste of deepfake actors who never achieve celebrity status or the ability to monetize their name and likeness.
Incorporating Celebrity Deepfakes in Digital Content
Individuals have also leveraged celebrity deepfakes on social media platforms, further highlighting the pervasiveness (and accuracy) of the underlying technology. In early 2021, a Belgian digital AI artist worked with a Tom Cruise impersonator to generate very realistic videos of “Tom Cruise” on TikTok under the account @deeptomcruise. Those videos featured “Tom Cruise” partaking in quirky activities, from falling and telling a Soviet Union joke in a retail store to performing industrial clean-up services, and attracted hundreds of thousands of views. Also, a deepfake of Harry Styles demanding more strawberries in a musical ode to his song Watermelon Sugar went instantly viral on TikTok last year.
If an individual or business would like to create a celebrity deepfake for media content, it should carefully consider with an attorney whether it is permitted to do so under applicable law. It should navigate some key legal bases to post that type of content, including whether the content is a protected class of free speech (e.g., a parody), whether the celebrity’s rights of publicity have entered into the public domain and whether it has a fair use defense to a copyright infringement claim. Otherwise, as in all other cases, consent is likely required for use of the talent’s likeness in this context.
Considering State Laws
An individual or business should also consider recent state laws that specifically address synthetic and digitally manipulated media.
For example, in November 2020, New York enacted a law that expressly bans the use of “a deceased performer’s digital replica” in audio-visual content, for 40 years after the performer’s death, if that use is “likely to deceive the public into thinking it was authorized.” This could prohibit the use of deepfakes in instances such as the Anthony Bourdain documentary Roadrunner. There, controversially, the film’s director leveraged deepfake technology to generate three lines that brought Bourdain’s “voice back to life” in order to complete the production following his death, despite the celebrity chef’s widow, Ottavia Bourdain, asserting that she did not give permission for such use.
On the political front, Texas enacted a law in September 2019 that banned disseminating deceptive “deepfake videos” intended to damage candidates or influence a voter base within 30 days of an election. The following month, California passed a similar law but specified that the period at issue is within 60 days of an election. Further, the platforms that host deepfakes will also need to consider compliance concerns regarding claims of deception.
Augmenting Video Game Characters with Deepfakes
The gaming industry is another natural arena for disruption by deepfakes, particularly with respect to avatars. A key premise of many games is a player assuming the role of a character, such as Luke Skywalker or Princess Leia from Star Wars. But an even more immersive gaming experience would be not simply controlling Luke or Leia with a gamepad, but also having the avatar track your face and mouth movements—something deepfake technology is making a reality. Further, with deepfake-generated synthetic speech, it is possible to make your voice sound like Luke or Leia, and this has sometimes resulted in unanticipated positive consequences. For example, these so-called “voice skins” are enabling LGBT people to change their in-game voices, resulting in more pleasant gameplay—an unsurprising finding given the 2020 statistic from the Anti-Defamation League that more than half of voice chat users are harassed during gameplay, and 37% of LGBT players are harassed on the basis of their sexual orientation.
Of course, general purpose technology like this also has the potential to be misused, such as for fraudulent impersonation for financial gain or fraudulent logins of voice-gated systems. And deepfake technology will impact nonplayer characters (NPCs) as well as your own avatar. The combination of impressive natural language generation models such as GPT‑3 paired with gaming deepfakes will result in NPCs possessing the limitless ability to converse with your avatar with convincing synchronized face and mouth movements without needing to follow specific scripts. Video game developers will need to analyze their existing licensing arrangements with the content owners of these characters and story arcs to determine whether the deepfake use cases are permitted.
Other Potential Benefits
In addition to the economic benefits of using deepfakes discussed above, the underlying technology can also be used for social good in digital media. For example, an HBO documentary entitled Welcome to Chechnya details the lives of LGBT activists forced to live in secrecy under threat of execution. To protect the identities of these activists, the documentary used deepfake wrappers, where the director reviewed only wrappers who were themselves LGBT activists but resided in countries free from the threat of death due to their sexual orientation. Deepfakes have also been used to create unique and bespoke voices for the millions of people who rely on synthetic speech, and have been used to create unique and bespoke voices for the millions of people who rely on synthetic speech to communicate.
Practical Considerations Going Forward
As deepfakes continue to permeate various facets of digital media, individuals and businesses seeking to leverage the underlying technology will have to preemptively think through their existing contractual arrangements and navigate applicable law on this topic. Further, individuals who enter into talent agreements should carefully review the terms regarding their rights of publicity to ensure that they have sufficient control in how those rights might be used in conjunction with artificial intelligence-based technologies. If approached thoughtfully, the development and use of deepfakes can be leveraged to promote social good, both commercially and socially.
By John Kind
Nearly 25 years ago, the Digital Millennium Copyright Act was added to the Copyright Act. Among its provisions were “paracopyright” measures extending protection in areas well beyond that of traditional copyright law. One such measure is found in § 1201 of the Copyright Act, which prohibits a person from circumventing a technological measure that effectively controls access to a copyright-protected work.
Recent Anti-circumvention Exemptions
An early example of § 1201 enforcement arose when Princeton University Professor Edward Felten and his information technology research team proposed to present their paper on digital music access control technologies at a conference in 2001. The Recording Industry Association of America (via the Secure Digital Music Initiative) argued that even discussing technology that could bypass such controls could be illegal under the DMCA, resulting in a short-lived but highly publicized lawsuit. These provisions of the copyright law can seemingly resemble patent law, in the sense that they seek to regulate certain uses of copyrighted works.
In part because this expansion might have unintended negative consequences, the DMCA included a provision that the Librarian of Congress should, every three years, determine appropriate exemptions from these anti-circumvention provisions for certain classes of works. The most recent rules on these exemptions became effective on November 29, 2001. The “Eighth Triennial Proceeding” to update the exemptions continues on many prior exemptions, but there are some extremely important proposed updates to the exemptions. For example, one exemption allows open-source authors to investigate devices with embedded open source software to check for license compliance.
The mechanism for these exemptions is as follows: the Register of Copyrights conducts a formal public process, consults with the U.S. Department of Commerce (the Executive Branch counterpart to the Library of Congress for IP issues), and makes recommendations to the Librarian of Congress. Throughout the past 20 years or so, exemptions through such rulemaking have been made for various classes of works where, per the legislative history replete with multiple negatives, “the lack of an ability to waive the circumvention prohibition might undermine the fair use of copyrighted works.”
While the DMCA itself includes certain permanent exemptions (e.g., for certain library browsing), other exemptions were to be revisited every few years to ensure that the technologies and their uses warranted the exemptions. Thus, in June 2020, the Copyright Office issued a notice of inquiry requesting petitions for renewal of existing exemptions, petitions in opposition to renewal and petitions for new (or expanded) exemptions. In response, it received 32 renewal petitions and 26 petitions for new/expanded exemptions. In October 2020, the Office issued a Notice of Proposed Rulemaking identifying the existing exemptions for which the Register intended to recommend renewal, as well as proposed classes for new exemptions. This invited several rounds of public commentary. Overall, the Office received 161 total submissions in response to the October 2020 Notice. After further public hearings in early 2021 and additional public input, the Register provided her recommendations to the Librarian of Congress on October 19, 2021. (The recommendations document is quite detailed, spanning more than 350 pages.) On October 28, 2021, the Librarian of Congress Adopted final regulations based on the recommendations.
New Exemptions Adopted in 2021
On the renewal side, not much changed. The final rule resulted in renewals of all of the existing exemptions, in some cases using those as the baseline in assessing whether further expansion was appropriate. These included things such as educational uses of audiovisual works, e.g., the use of motion pictures for massive open online courses and captioning/audio descriptions of works for students with disabilities. For existing exemptions relating to software, they included uses relating to patient data on networked medical devices; unlocking cellphones/jailbreaking; repair of motorized land vehicles (including tractors); repair of smartphones, home appliances and home systems; security research; software/video game preservation; and operation of 3D printers to allow use of alternative feedstock.
More notable were areas of new or expanded exemptions. A new exemption that is of particular interest to technology-based companies allows for circumvention of technical measures on any device or machine (except a gaming console) to investigate potential violations of licenses for “free and open source software” (FOSS). Any such investigation must be performed on a legally obtained device and not violate other laws. The investigation must also be performed by, or on behalf of, a party with standing to bring a breach of license or copyright claim that has a good-faith belief that the investigation is necessary. Further, the investigation must be for the sole purpose of investigating potential infringement. However, how this exemption will work in practice is not yet clear.
Given the prevalence of FOSS code, there will often be good reason to believe that it is present in any given device. Without inspecting the code, it may be challenging or impossible to know whether any given piece of FOSS code is present. It could be plausibly argued that the mere fact that a particular piece of FOSS code would be useful in a device is sufficient to establish good faith grounds to investigate. In contrast, one could argue that building a good faith argument requires at least some independent evidence of use of the FOSS code, which in many cases may be impossible to obtain without circumventing the technical measures protecting the code. Furthermore, once an investigation has been conducted and source code has been exposed, it is extremely difficult—if not impossible—to control how knowledge of that code is used. There are obvious examples of misuse of the exemption (e.g., publishing proprietary code online) that may be policed, but there are many gray areas that will need to be evaluated on a case-by-case basis.
Another new exemption of interest allows for circumvention of technical measures on motion pictures and literary works stored on DVDs or Blu-ray discs, or made available for digital download, for researchers to perform text and data mining. The circumvention must be applied by a researcher affiliated with a nonprofit institution, or a student or information technology staff member under such a researcher’s direction, and the institution must legally own or license a copy of the copyrighted work. Further, the work may only be watched or read for the purpose of validating the results of the research and the institution must employ effective security measures to prevent further dissemination of the work. Exactly what constitutes “effective security measures” should be determined by an agreement between the institution and interested copyright holders. Absent such an agreement, the institution must use the same measures that it uses to keep its own highly confidential information (e.g., government-issued identification numbers, medical or health insurance information, and payroll systems) secure. This exemption has limited scope, but for institutions and research projects that qualify, it may provide easier access to rich sources of data for analyzing cultural phenomena and development.
Proceed with Care
The new exemptions expand the situations under which individuals may circumvent technical measures protecting copyrighted material. The new exemption for investigating possible copyright violations has significant ramifications for both copyright holders and users of FOSS code. Copyright holders now have a powerful tool for ensuring that users comply with the terms of the license under which the code is distributed.
However, copyright holders should be aware of the limitations and, ideally, have detailed procedures and records to ensure they can demonstrate a good faith reason for all investigations and that anything learned through those investigations is not misused. For example, copyright holders should consider deleting the majority of source code extracted from devices and storing only conclusions and short extracts necessary to prove license or copyright violations. Similarly, users and manufacturers of computing devices should be aware of the increased risks of third parties accessing any code included on those devices. To minimize this risk, including detailed documentation of all FOSS code used by a device is advisable to undercut arguments that there is a good faith case to investigate the code used by the device.
Yes, you read that correctly. Trademark practice at the U.S. Patent and Trademark Office as we know it is about to change (well, at least in part).
The USPTO recently published its final rule implementing certain provisions of the Trademark Modernization Act of 2020. The TMA was enacted to combat the recent uptick in fraudulent filings (a sizeable share coming from China) and restore integrity to the registry. As a result, several key changes to the trademark rules of practice went into effect in December 2021, with more to come in December 2022. According to the USPTO, “Individuals, businesses, and the USPTO [will] now have new tools to clear away unused registered trademarks from the federal trademark register and the USPTO [will have] the ability to move applications through the registration process more efficiently.”
Here is a quick rundown of the key changes:
The full rule, which is relatively easy to follow, is available here, and the USPTO’s new Trademark Modernization Act webpage, with helpful information in Q&A format, can be found here.
By Diana Buck
The traditional understanding in copyright law is that the concept of “prior art” is only applicable to patents and that the term is not relevant in assessing whether a defendant has infringed someone’s copyright. Patent law demands that an invention must be new and novel to receive protection. Prior art, which can be broadly described as any preexisting knowledge in the world that’s available to the public, is frequently used by defendants in litigation to argue that a plaintiff’s so-called invention should never have received patent protection because the plaintiff's invention was not new.
In copyright, however, a work of authorship does not need to be new; it need only have some spark of creativity that came from the author in order to receive protection. For example, it does not matter if someone’s photo of the Eiffel Tower looks just like the innumerable similar images on the internet; it just matters that impermissible copying did not occur. Nevertheless, in assessing what exactly is unique to the author and thus is protectible under copyright law, courts will filter out all unprotectable elements in a work. The U.S. Court of Appeals for the Ninth Circuit uses a two-part extrinsic-intrinsic test to determine whether a defendant has impermissibly copied the protected elements of a work. If a defendant has copied only unprotected elements of a work, then the plaintiff has no claim.
Yet there’s a trend developing in district courts to refer to “prior art” in copyright cases when assessing which parts of a plaintiff’s work are protectable. In Johannsongs-Publishing v. Lovland, Johannsongs alleged that the song, “You Raise Me Up,” written by defendants Rolf Lovland and Brendan Graham, impermissibly copied from the old Icelandic hit song, “Söknuður.” In a motion for summary judgment at the district court level, the defendants argued that the similarities between the songs were not sufficient to find infringement and actually derived from a song in the public domain and “several other prior art songs.”
Both parties submitted expert reports, yet the district court found only one, from the defendants’ expert, that properly filtered out all unprotectable elements, including prior art songs. The plaintiff’s expert reports had failed to conduct a prior art investigation, so, according to the district court, their comparison of the songs failed to “filter out unprotectable prior art elements, which is the foundation of the extrinsic test.” Thus, the court granted the defendants’ motion for summary judgment, and the plaintiff appealed. In November 2021, the Ninth Circuit issued a fairly brief decision on that appeal, holding that the district court properly excluded the plaintiff’s expert reports because they “failed to filter out similarities that are attributable to prior art, as required under the extrinsic test.”
What does all this mean for future copyright litigants?
Neither the district court nor the Ninth Circuit explained what they meant by the term “prior art.” It may simply have been a shorthand to refer to examples of other artistic works, and the courts’ intention was to perform a normal copyright infringement analysis. Or, it could signal that courts are beginning to use prior art to assess whether a work deserves wide or narrow protection instead of looking to more traditional doctrines of unpredictability, like the idea-expression distinction or the concept of scènes à faire.
The bottom line is that, whether or not courts using the term “prior art” in referring to other artistic works is merely semantics, litigants should make sure that they conduct a prior art search and be prepared to argue how a work is or isn’t different from the rest.
On November 1, 2021, the Director of the U.S. Patent and Trademark Office granted the first “Director review” to Samsung SDI Co. Director review is a new interim procedure that allows a party to seek review of a final written decision of the Patent Trial and Appeal Board by the Director of the USPTO. Here, the Director vacated the Board’s final written decision invalidating patent owner Samsung’s lithium-ion battery patent. The decision marks the first time the Director has exercised this new authority following the U.S. Supreme Court’s decision in United States v. Arthrex.
The Director review procedure is, in fact, borne out of Arthrex, in which the Supreme Court held that for the appointment of the Board’s administrative patent judges to be constitutional under the Appointment Clause, decisions rendered by the APJs must be subject to review by the Director. The Court held that the unreviewable authority wielded by APJs during inter partes reviews is incompatible with their appointment by the Secretary of Commerce to an inferior office, and that only an officer properly appointed to a principal office may issue a final decision binding the Executive Branch in an IPR. It was following the Arthrex decision that the Director implemented the new interim procedure for Director review, wherein a party can request a Director review by filing a request for rehearing and notice of the request.
In the recent matter, Samsung filed a request for Director review, challenging the Board’s decision invalidating its patent in an IPR filed by Ascend Performance Materials Operations, or APM. The Samsung patent is directed to an electrolyte composition including a nitrile compound additive to increase a lithium-ion cell’s thermal impact durability. It issued from a provisional application filed on September 7, 2012, and a non-provisional application filed on March 14, 2013. APM argued that the claims were not entitled to the September 7, 2012 priority date due to lack of written description support in the provisional application. APM then argued that the claimed nitrile compound additive was anticipated by an NEC patent application that published on March 8, 2012, more than one year before the non-provisional filing date, thus qualifying as prior art under 35 U.S.C. § 102(b). The Board agreed and found the claims anticipated in a final written decision.
Samsung argued in its request for Director review that patent claims are awarded priority on a claim-by-claim basis based on the disclosure and that the Board failed to address the priority date of dependent claims 5 and 17. The Director concurred and ordered the Board to issue a new final written decision to address whether dependent claims 5 and 17 were entitled to the provisional application’s date, as well as the patentability of claims 5 and 17, in view of the appropriate priority date.
The Director review procedure opens up a new option for parties to challenge the outcome of a final written decision. Many uncertainties remain, such as whether a Director review is more favorable than a request for rehearing to the Board or an appeal to the Federal Circuit. Answers to these questions will only emerge over time. The one current certainty is that the Director review process introduces yet another layer of complexity that must be factored into a party’s enforcement or defense calculus.
Most lawyers know, at least generally, that IP infringement and misappropriation actions are subject to various statutes of limitations. Patent actions need to be brought within six years, copyright actions within three, and so forth. As with so many other IP issues, the devil is often in the details. The issue we’ll focus on here is specific to cases involving employees who depart one company for a competitor.
In such situations, employers typically will notify both the employee and the new company of the employee’s ongoing obligations of confidentiality to the old employer. Exit processing may also involve examination of the employee’s computer/network activity in the weeks leading up to their departure, particularly where the circumstances suggest possible misappropriation. However, the old employer has far less insight into what the new employer may be doing with information from the employee.
Such was the situation in a recent case (Schwan’s Company v. Cai) involving two food manufacturing companies and an employee leaving one to join the other. The employee, Cai, was a research scientist working on food ingredient technologies for the plaintiff, Schwan’s. On November 8, 2017, the scientist was offered a job by the defendant, Conagra, and accepted it two days later. He only submitted his letter of resignation on December 18, 2017, requesting an effective date of the following January 5. Cai’s supervisor immediately asked if he intended to move to a competitor, and he allegedly denied that he did. However, Schwan’s apparently learned later that day that Cai was in fact going to Conagra, so he was immediately terminated. Just nine days thereafter, Cai filed two U.S. patent applications (and one corresponding Chinese patent application) relating to work he had been doing for Schwan’s.
In January 2018, Schwan’s sent a letter informing Conagra that Cai was in possession of Schwan’s confidential information and asking that Cai not use this information for Conagra’s benefit. In October 2020, Schwan’s brought suit against Cai but not Conagra. Based on discovery in that suit, Schwan moved to amend its complaint to include Conagra as a defendant as well. Conagra sought to dismiss the claims against it based on the three-year statute of limitations applicable to the federal and state trade secrets acts.
The district court undertook a comprehensive analysis in looking at the statute of limitations issues. Initially, it observed that a motion to dismiss is typically “not the proper stage to resolve whether a claim is barred by a statute of limitations.” However, “if the complaint itself shows that the claim is time-barred,” dismissal may be appropriate. In this case, both the state and federal trade secrets acts mark the beginning of the period as when misappropriation is “discovered or . . . should have been discovered.”
Conagra argued that Schwan’s met this standard in early 2018 based on what it learned about Cai. The court, however, denied Conagra’s motion, saying that knowledge of the employee’s behavior did not serve to inform Schwan’s about Conagra’s behavior in any way. The court also emphasized that questions of fact often make limitations issues unsuitable for resolution on summary judgment, much less a motion to dismiss.
Adding further complexity to this issue, another very recent decision from a federal court in Azima v. Del Rosso, looking at a fairly similar fact pattern, dismissed a federal trade secrets count but not a corresponding state trade secrets count. The federal Defend Trade Secrets Act includes language making clear that the limitations period begins when the plaintiff discovers the first act of misappropriation that continues over time. On the other hand, the North Carolina statute is silent on the point, and under that state’s general “continuing wrong” doctrine, the applicable period restarts when an allegedly wrongful act is repeated.
These cases serve to underscore lessons for both the old employers and the new employers. For companies losing an employee to a competitor, it’s important to realize that even slight nuances in how the facts arise (and how they are presented in a complaint) can determine when the limitations clock begins to run with respect to the competitor—and that may differ from the time frame applicable to the departing employee. For companies hiring from a competitor, it is important to keep in mind that potentially differing time frames may subject the company to a lawsuit even after such a claim directly against the employee might be barred.
For these reasons, among many others, companies are well served by taking both their employee inbounding and outbounding diligence duties seriously.