On a Friday morning in October, within the lobby of a modern San Francisco high-rise, Matthew Butterick found himself on his way to the elevators when a security guard intercepted him. Politely, the guard inquired if he was lost.
This encounter stemmed from a simple misunderstanding. Clad in checkerboard Vans, a black baseball cap, and a windbreaker, Butterick hardly fit the mold of a typical corporate professional. In fact, he seemed more inclined to poke fun at the traditional corporate archetype. With equal politeness, he clarified that he was, indeed, a lawyer with legitimate business in the building. His colleague, Joseph Saveri, helmed an antitrust and class-action firm with its headquarters located there.
Although his appearance may suggest otherwise, Butterick stands as an unexpected leader behind the initial surge of class-action lawsuits against major artificial intelligence corporations. His mission revolves around ensuring that writers, artists, and other creatives retain authority over the utilization of their work by AI systems.
This trajectory was not what Butterick had envisioned for himself. Until recently, he had not actively practiced law, nor does he harbor any anti-technology sentiments. For the better part of his life, he engaged in independent design and programming work, delving into specialized software projects. “I’m just a guy at home,” he remarks casually. “No assistants, no team.” His idea of enjoyment? Crafting an application from scratch for personal use. While he does make appearances in the Bay Area for court obligations—given that all the lawsuits are filed in the Northern District of California—he primarily operates from his Los Angeles residence, where he resides with his wife.
However, with the rise of generative AI, he revived his dormant law degree specifically to combat this cause. Presently, he collaborates with Saveri as co-counsel on four distinct cases, commencing with a lawsuit initiated in November 2022 against GitHub. The lawsuit alleges that the AI coding tool, Copilot, from the Microsoft subsidiary, infringes upon open-source licensing agreements. Subsequently, they represent a spectrum of programmers, artists, and writers, including comedian Sarah Silverman, who assert that generative AI entities are encroaching upon their rights by utilizing their work for training without consent.
While each complaint adopts slightly varied legal strategies, collectively, they embody a movement to empower creative individuals to dictate the usage of their work in AI training. “It’s a form of resistance,” Butterick articulates. This endeavor poses a direct challenge to AI companies, portraying the training methodologies of their tools as fundamentally unethical. Even within the realm of copyright and intellectual property, many scholars view this initiative as a long shot.
The pivotal question remains: How will the judicial system perceive these endeavors?
These legal battles stand out as some of the most closely monitored disputes of the present time. The advent of the AI era has sparked a renaissance in Silicon Valley, following a decade marked by escalating public concerns regarding technology’s impact on society. The fervent enthusiasm surrounding tools like ChatGPT has ushered in a new era akin to the Second Age of “Move Fast and Break Things.” Amidst the hype and staggering valuations—such as OpenAI’s reported value of $80 billion—generative AI distinguishes itself by its tangible utility. While it undoubtedly mirrors a gold rush, unlike past cycles revolving around the metaverse and cryptocurrency, generative AI delivers practical value. The industry acknowledges the solidity of its foundations. However, these lawsuits, which allege legal transgressions by OpenAI, Meta, Stability AI, and other companies in the development of their tools, pose a significant threat to the burgeoning momentum of the generative AI movement. The stakes have been raised to unprecedented heights.
The outcomes of these legal confrontations possess the potential to solidify the industry’s existing landscape or catalyze radical transformations. While a security guard may not have recognized Butterick, the legal teams at AI enterprises have undoubtedly taken notice. Their fates may hinge on the efficacy of his arguments.
Hailing from New Hampshire, Butterick emerged as a standout student, securing admission to Harvard in the late ‘80s. However, amidst his peers’ conventional aspirations, already contemplating law school, he gravitated towards a more esoteric domain. Nestled in the basement of his Cambridge dormitory, the Bow & Arrow Press operated a workshop, offering students a unique platform to delve into traditional printing techniques. This quaint, cherished space, adorned with whitewashed walls plastered with posters and antiquated machinery, attracted unconventional aesthetes. Encountering this environment proved transformative for Butterick.
His fascination with typography burgeoned, leading him to delve into font design while still in school. “People in my life thought it was a ridiculous thing to do,” he recalls. Delighting in the manipulation of archaic tools, his true passion lay in devising innovative typefaces. Following his graduation in 1992, he harbored aspirations of exploring the burgeoning tech landscape in San Francisco, a city ripe with opportunities for a visionary intent on modernizing typography for the digital age. Thus, two years later, he embarked on his westward journey.
As fate would have it, lawyers harbored a profound appreciation for fonts.
In alignment with numerous Ivy League transplants venturing to the Bay Area with aspirations of leaving a mark in the tech realm, Butterick ventured into the startup domain. Fondly reminiscing about his “dotcom adventure,” he founded Atomic Vision, a web design company. By the age of 28, he oversaw a team of approximately 20 employees. However, the managerial aspect failed to resonate with him. Thus, when an opportunity arose to sell the company in 1999, he seized it.
Endowed with financial security and grappling with uncertainty regarding his next pursuit, Butterick opted to tread the path of countless young adults navigating life’s uncertainties: he enrolled at UCLA to pursue a law degree. Post-graduation, he initiated a website titled Typography for Lawyers. Initially conceived as a niche endeavor, the website swiftly gained traction. This momentum culminated in the publication of a remarkably popular book under the same title in 2010. His typefaces found widespread adoption in courts and private firms across the nation. Following the integration of his Equity font, a Fifth Circuit judge lauded it as a fully-equipped F-150 in contrast to the Buick that was Times New Roman. “The stuff of finicky opinion-readers’ dreams,” the judge extolled.
Joseph Saveri emerged as another font enthusiast within the legal sphere. “These AI cases would not be happening if we weren’t mutual admirers of typography,” Butterick notes. Their interaction commenced in 2012 when Saveri sought advice on formatting pleadings. Saveri undertook cases that commanded Butterick’s admiration, such as a triumphant class action against Meta on behalf of content moderators.
Occasionally perusing Butterick’s eclectic personal blog—a fusion of gift recommendations and acerbic reflections on tech and typography—Saveri stumbled upon a fervent post authored by Butterick shortly after GitHub unveiled its Copilot AI tool in the summer of 2022. The post vehemently argued that the tool pilfered programmers’ work, resonating with a broader community of discontented individuals. Subsequently, Saveri extended an invitation for collaboration. Butterick was not the sole dissenter perturbed by Copilot; however, his possession of a dormant law degree offered him a unique platform to effect change. In November 2022, they lodged their inaugural complaint.
Concurrently, as Butterick and Saveri laid the groundwork for their lawsuit against GitHub, artist Karla Ortiz embarked on her personal AI revelation. Hailing from her loft in San Francisco, she delved into the intricacies of image generator training methods. Disconcerted by her findings, she articulated her sentiments. “It all felt really exploitative and gross,” she remarked. “It’s existential. I feel there’s no choice but to stand up and be loud.”
Ortiz’s creative repertoire spanned diverse disciplines, encompassing fine art alongside conceptual art and illustration for entertainment and gaming entities like Marvel, Ubisoft, and Universal. For working artists, projects of this nature often serve as financial sustenance, yet they also represent the very projects imperiled by the AI revolution. Ortiz bore witness to the decline of pitch work, a realm where artists are commissioned to craft visuals facilitating idea pitches to investors. “That work was bread-and-butter for so many of us,” she lamented. “It’s completely gone.”
In collaboration with a cohort of like-minded individuals, Ortiz initiated outreach to legal professionals in pursuit of representation for artists desiring to litigate against AI corporations infringing upon their rights. Rebuffed by a cascade of rejections, she chanced upon the Copilot case. Reaching out to Saveri’s firm, she encountered receptivity instead of dismissal. Ortiz discerned a profound understanding from Butterick regarding her predicament. “He understands artists,” she affirmed. Recognizing a shared vision, she discerned that she had found her legal advocates.
In January 2023, Ortiz and two fellow artists emerged as the lead plaintiffs in Butterick and Saveri’s subsequent class action, directed against Stability AI, Midjourney, and DeviantArt. Subsequently, in July 2023, Butterick and Saveri initiated two additional class-action lawsuits, this time championing writers’ rights. These lawsuits targeted OpenAI in duplicate, slated for consolidation into a singular case, alongside a lawsuit against Meta. The duo’s endeavors began attracting attention. Ortiz testified before the US Senate Judiciary Subcommittee on Intellectual Property, engaging in a discourse on AI and copyright alongside domain experts. She emerged as a prominent advocate for artists, spearheading calls for modifications to AI training methodologies.
At this juncture, the backlash against generative AI’s impact on the arts had gained substantial momentum. As AI tools surged in popularity, concerns regarding their implications on creative industries burgeoned. In Hollywood, the Writers Guild of America and the Screen Actors Guild-American Federation of Television and Radio Artists embarked on concurrent strikes, advocating for regulatory frameworks governing AI utilization in their respective domains. A wave of discontent rippled across diverse creative sectors—ranging from filmmaking and podcasting to music and journalism—each grappling with the disruptive influence of AI. More than 15,000 writers, organized by the Authors Guild, endorsed a letter addressed to generative AI entities, urging compensation and a licensing framework for training data utilization.
“Publishers, authors, visual artists, software developers, the recording industry—everybody’s angry,” remarked Pam Samuelson, a copyright law expert affiliated with the Berkeley Center For Law & Technology. “Some other technologies have been upsetting to one group of people. Now, everybody’s upset.”
Preceding Butterick and Saveri’s initial lawsuits, the sole major case revolving around AI training data involved a lawsuit predating the generative AI surge. In 2020, media conglomerate Thomson Reuters filed suit against an artificial intelligence research firm named Ross Intelligence. Apart from owning the news agency Reuters, Thomson Reuters also commands the legal research entity Westlaw. The lawsuit alleged that Ross Intelligence trained its AI tool on Westlaw’s legal summaries sans appropriate licensing. This dispute is slated for trial next spring, potentially serving as the pioneering precedent-setting case in this domain.
Although the Thomson Reuters case appeared as an isolated incident, AI corporations now confront the stark reality of heightened vulnerability to litigation, as a barrage of lawsuits inundates the legal landscape.
At the onset of the year, Getty Images instigated legal action against Stability AI across both US and UK jurisdictions. Subsequently, in September, two additional cohorts of writers lodged complaints against OpenAI. The first group featured prominent figures like George R. R. Martin, John Grisham, and Jonathan Franzen, whereas the second group comprised individuals such as Michael Chabon, Ayelet Waldman, and David Henry Hwang. In October, a consortium of Christian authors, including former Arkansas governor Mike Huckabee, initiated another lawsuit. (“Imitation is the sincerest form of flattery,” quipped Saveri.) Concurrently, a cluster of music labels, including Universal Music Group, sued Anthropic for incorporating copyrighted materials in its training data, alleging unlawful distribution of lyrics in its outputs. Most recently, a legal firm representing a coalition of nonfiction writers filed a proposed class-action lawsuit against OpenAI and Microsoft.
With the filing of these initial lawsuits by Butterick, the trajectory has transitioned into a full-fledged IP War.
Matthew Butterick’s legal endeavors possess the potential to mold the trajectory of AI’s future.
Copyright law, at times, assumes a dormant posture. Periodically, however, transformative moments surface, catalyzed by new technologies that disrupt cultural norms. In the late 1890s, the advent of the piano roll epitomized such a juncture, enabling player pianos to generate music sans human intervention. Several experts, consulted for this narrative, underscored piano rolls as a quintessential case study at the intersection of copyright and technology. While piano rolls captivated audiences and offered pianists a novel medium to record their performances, they instigated trepidation among live performers, who viewed their careers as imperiled, and ire among music publishers, who litigated against piano roll manufacturers for copyright violations. (Does this narrative ring a bell?) They faced defeat. Presently, the outcome remains uncertain. Nonetheless, copyright experts view this as a formidable battle, unlikely to culminate neatly or devoid of friction.
“At present, we’re embroiled in a quintessential legal dispute within a domain of emerging technology, where stakeholders grapple with metaphorical interpretations,” observes James Grimmelmann, a Cornell University professor specializing in internet law. “It’s about narrating a compelling account of AI’s operational dynamics.”
For AI corporations, this translates into leveraging narratives that portray their tools as merely training on art, akin to human apprentices studying under the tutelage of master painters. When AI entities articulate their training methodologies, they often draw parallels with artistic learning processes. (GitHub, Meta, OpenAI, and Stable Diffusion declined to comment on this narrative.) Conversely, their adversaries depict AI training as a mechanized menace—depicting algorithmic extraction camouflaged as innovation—a portrayal of corporations mechanizing the act of art theft.
Grimmelmann commends Butterick and Saveri for their strategic storytelling. “Among the best-formatted complaints I’ve encountered,” he lauds. In his view, they have tapped into the emotional turmoil experienced by creative individuals upon realizing that these cutting-edge tools are founded upon their labor, often generating content that competes with their creations. Nonetheless, they have encountered their initial stumbling blocks.
“I’m just one piece of this—I don’t want to call it a campaign against AI, I want to call it the human resistance.”
In October, US District Judge William Orrick pared down the Stability AI artists’ case, dismissing several claims. Nevertheless, he granted the plaintiffs an opportunity to amend their claims, affording them the chance to refine their arguments before resubmission later this year. In November, a similar scenario unfolded in the Meta case. US District Judge Vince Chhabria indicated his inclination to uphold Meta’s motion to dismiss allegations concerning copyright violations by text generated through its AI, Llama. However, as Butterick notes, the judge hinted at the plaintiffs’ ability to amend their claims. Notably, Meta did not contest the central claim in the case—that the utilization of training data in itself constitutes an infringement—and thus, this aspect progresses. “Thus far, it appears that they have not convincingly persuaded judges to fully embrace their narrative. However, the judges remain receptive,” Grimmelmann observes.
Judge Chhabria explicitly expressed his skepticism regarding the claims he dismissed in their current form. He acknowledged their core theory but found the auxiliary theories confounding. “Your remaining theories of liability I don’t understand even a little bit.”
He particularly critiqued the notion that Llama—the tool in its entirety—is an infringing entity. “That makes my head explode when I try to understand that.”
Several copyright experts echo this skepticism. Berkeley’s Samuelson, for instance, deems it “absurd” to assert that everything outputted by an AI is inherently infringing due to its training process. Moreover, she exhibits scant patience for arguments positing copyright law as a shield against evolving creative landscapes. “Copyright isn’t a jobs program.”
Other objections surface as well. The US Copyright Office is presently receiving comments for its AI study, unveiling starkly contrasting perspectives on whether AI training contravenes copyright regulations. The Copia Institute, a think tank founded by Techdirt’s Mike Masnick, bluntly contends that artists lack the prerogative to restrict their work from AI training, deeming said training akin to reading rather than copying. “Ultimately when we speak of training AI we are speaking about letting software ‘read’ or otherwise consume works on behalf of the people developing it. Copyright law does not prevent reading,” their submission outlines. Should the courts embrace this perspective—portraying training as consumption—it will be arduous to sway them with arguments predicated on training as theft.
Most experts refrain from prognosticating the precise outcomes of these cases. Nonetheless, there is a consensus regarding the likely defense strategy of AI corporations: the fair use doctrine, a prevalent defense mechanism against copyright infringement allegations. Some defendants have already alluded to this strategy; for instance, Midjourney cited “obvious fair use defenses” in its motion to dismiss Ortiz’s case.
In the US context, “fair use” denotes a scenario where copyright infringement transpires but is permissible to foster creativity and freedom of expression. It embodies a broad, ambiguously defined, yet indispensable doctrine. It underpins the ability of satirists to publish parodies of literary works and films, enables scholars to incorporate excerpts in their research, and legitimizes personal recording of televised content for subsequent viewing.
Big Tech has embraced fair use wholeheartedly. In 2013, confronted with a lawsuit from the Authors Guild, Google successfully contended that copying millions of books for online snippets was permissible, as it facilitated the creation of a searchable database for public access. (Subsequently, another judge rendered a similar verdict concerning a distinct book-scanning initiative, the HathiTrust Digital Library.)
Nonetheless, fair use remains a nebulous concept. “Early on, we heard from opponents that the Authors Guild v. Google case would be determinative,” Butterick reflects. If Google was granted leeway, why couldn’t they amass millions of books for their projects? He remains unconvinced. “The essence of the Google Books project was to direct users to the original books, right? ‘Here’s where you can find this book in a library.’ Generative AI fails to fulfill this function. It does not guide users to the original work. On the contrary, it competes with that work.”
He finds solace in a recent US Supreme Court ruling: Andy Warhol Foundation for the Visual Arts, Inc. v. Goldsmith. Although unrelated to AI, the case holds profound implications for ongoing litigations, demonstrating the judiciary’s willingness to interpret fair use in a more restrictive manner than in the past. One of the primary arguments advanced by AI critics revolves around the plight of human creators, who deserve agency in determining the outcomes of their narratives, particularly when AI tools leverage their creations, often generating content that rivals their original work. Warhol v. Goldsmith underscores the potential efficacy of focusing on the competitive aspect.
Presently, Butterick is engrossed in refining the complaints for the Stability AI and Meta cases, alongside preparations for the impending lawsuits against GitHub and OpenAI. This legal terrain may appear agonizingly sluggish. While the amended complaint in the Stability AI case is due this month, the judicial ruling is deferred until April, with discovery activities paused in the interim.
This sluggish pace poses a liability for the plaintiffs. “The passage of time has favored
