OpenAI, Inc. develops artificial intelligence software involving large language models (“LLM”) known as ChatGPT.  In 2023, several authors, including the comedian Sarah Silverman, filed putative class action lawsuits alleging various copyright infringement claims. On February 12, 2024, a District Court in the Northern District of California issued its Order and ruled on the OpenAI defendants’ motions to dismiss various claims in the two pending putative class action lawsuits.

The plaintiffs in the two lawsuits alleged similar claims against the OpenAI defendants, specifically asserting claims for: (1) direct copyright infringement; (2) vicarious infringement; (3) violation of section 1202(b) of the Digital Millennium Copyright Act (“DMCA”); (4) unfair competition under section 17200 of the California Business & Professions Code; (5) negligence; and (6) unjust enrichment.  The OpenAI defendants moved to dismiss all causes of action alleged by the author plaintiffs with the exception of the first cause of action for direct copyright infringement.  (It is unclear from the Court’s Order as to why the OpenAI defendants did not also move to dismiss this cause of action.)

The author plaintiffs alleged that OpenAI infringed on their published works by using these works to help train its LLM.  The plaintiffs alleged that OpenAI copied their published books, which are protected by copyright law, and used them in a training dataset for its LLM.  Plaintiffs also alleged that if a ChatGPT user asked it to summarize one of their books, ChatGPT would generally generate “accurate summaries of the book’s content and themes.”

The Court began by recognizing the general rules that govern motions to dismiss in federal court actions.  In essence, to survive such a motion, a plaintiff must plead “enough facts to state a claim to relief that is plausible on its face.”  In essence, a plaintiff must allege sufficient “factual content that allows the court to draw the reasonable inference that the defendant is liable for the misconduct alleged.”

The Court then turned its attention to the various claims beginning with the vicarious copyright infringement claim.  The Court noted that the Copyright Act “grants the copyright holder exclusive rights to (1) `reproduce the copyrighted work and copies;’ (2) `prepare derivative works;’ and (3) `distribute copies of the copyrighted work to the public’.”  However, the Court noted that “the mere fact that a work is copyrighted does not mean that every element of the work may be protected.”

To allege a valid copyright infringement claim, a plaintiff must show that: (1) that he or she owns a valid copyright; and (2) that the defendant “copied aspects of his [or her] work.”  The Court noted that the second prong of this analysis “contains two separate components: `copying’ and `unlawful appropriation’.”  Generally, a plaintiff can satisfy these elements by showing “that the defendant had access to the plaintiff’s work and that the two works share similarities probative of copying while the hallmark of ‘unlawful appropriation’ is that the works share substantial similarities.”  Importantly, the Court noted that “a claim of vicarious infringement requires a threshold showing of direct infringement;” specifically, that “the defendant has: (1) the right and ability to supervise the infringing conduct; and (2) a direct financial interest in the infringing activity.”

The OpenAI defendant sought to dismiss the vicarious infringement claims on the grounds that: (1) plaintiffs did not allege that direct infringement occurred; (2) that there was no allegation that the OpenAI defendants had the right and ability to supervise; and (3) there was no allegation that the OpenAI defendants had a “direct financial interest.” 

The author plaintiffs argued that “because defendants directly copied the copyrighted books to train the language models,” they did not need to show a substantial similarity between the works, relying on the 2012 Ninth Circuit case Range Road Music, Inc. v. East Coast Foods, Inc., 668 F.3d 1148.  The Court noted that the plaintiffs apparently misunderstood the holding in Range Road because the court there excused the plaintiffs from having to show “substantial similarity because the infringement was the public performance of copyrighted songs at a bar.”  In furthering this reasoning, the Court noted that the author plaintiffs had “not alleged that ChatGPT outputs contained direct copies of” their copyrighted books and therefore they had to allege a “substantial similarity between the outputs [of ChatGPT] and the copyrighted materials.”  The Court granted the motion to dismiss but gave the author plaintiffs leave to file an amended complaint to try to satisfy the “substantial similarity element.”

The Court then turned its attention to the DMCA claim which in essence prohibits the intentional removal or alteration of copyright management information (“CMI”).  The Court recognized that to state a claim under this portion of the DMCA, plaintiffs must allege “what the removed or altered CMI was” and they must “show the requisite mental state as each of the three forms of section 1202(b) violations require `knowing or having reasonable grounds to know that [intentionally removing CMI] will induce, enable, facilitate or conceal’ infringement.” 

The plaintiffs alleged that the OpenAI defendants had “by design” removed CMI from their copyrighted books during the LLM training process.  The Court noted; however, that there were no specific factual allegations to support this claim.  Rather, the complaints included excerpts from ChatGPT outputs which included “multiple references to plaintiffs’ names suggesting that OpenAI did not remove all references to `the name of the author.’”  The Court continued by recognizing that even if the plaintiffs could show that the OpenAI defendants had knowingly removed CMI during the training process, they had not alleged “how omitting CMI in the copies used in the trainings set gave defendants reasonable grounds to know that ChatGPT’s output would induce, enable, facilitate or conceal infringement.”  The Court rejected the author plaintiffs’ arguments that OpenAI’s refusal to state which books it was using to train its models would deprive ChatGPT users from knowing “if any output is infringing.”  The Court noted that there was no caselaw to support this position and ruled that the claim under section 1202(b)(1) would be dismissed with leave to amend.

The Court next addressed section 1202(b)(3) of the DMCA which plaintiffs allege was violated because ChatGPT created outputs that were “derivative works” of the authors’ books and distributed those outputs without including the CMI.  The Court began by noting that “while it may be unlawful to recreate another’s work (e.g., under the Copyright Act), this conduct does not necessarily implicate the DMCA.”  The Court found it significant that the plaintiffs had not alleged that the OpenAI defendants had “distributed their books or copies of their books,” but rather, only that the “output from the OpenAI language models is an infringing derivative work.”  Because they had not alleged that ChatGPT reproduced their copyrighted works without CMI, the Court dismissed this portion of the DMCA claim but gave the plaintiffs leave to amend to allege additional facts.

The Court then turned its attention to the various claims brought under California state law, such as unfair competition and negligence.  As to the unfair competition claims, the Court recognized that “a business act or practice may violate the UCL if it is either `unlawful,’ `unfair,’ or `fraudulent’.”  The Court quickly dismissed that Plaintiffs had alleged claims under either the unlawful or fraudulent prongs in that as discussed above, plaintiffs had failed to allege unlawful conduct under either vicarious copyright infringement or violation of the DMCA; and likewise, there were no allegations in the complaint that the OpenAI defendants had engaged in fraudulent conduct.  With regard to the unfair prong, the Court noted that “one test for determining an `unfair’ practice is [whether] the gravity of the harm to the victim outweighs the utility of the defendant’s conduct.”  The Court noted that because the plaintiffs had alleged that the OpenAI defendants were using their copyrighted works to train their LLM for commercial profit, this conduct could constitute an unfair business practice.  Thus, the Court allowed this portion of the unfair competition claim to proceed.

With regard to the negligence claim, the Court found that plaintiffs had not alleged that the OpenAI defendants owed them any legal duty under negligence law with regard to “the control of plaintiffs’ information in their possession.”  The Court found that plaintiffs had not explained “how merely possessing their books creates a special relationship” and distinguished those cases where a negligence claim could be alleged against a defendant who was the custodian of a plaintiff’s personal or confidential information.  Like the other claims, the Court granted plaintiffs leave to amend their negligence claim to allege facts supporting a legal duty.

Finally, the Court turned to plaintiffs’ unjust enrichment claim and likewise found that plaintiffs had not alleged “that OpenAI unjustly obtained benefits from plaintiffs’ copyrighted works through fraud, mistake, coercion or request.”   Therefore, the Court dismissed the unjust enrichment claim, but once again gave the plaintiffs leave to amend.

Curiously, the Court noted that the OpenAI defendants had not raised the issue of whether their claims for unfair business practices, negligence and/or unjust enrichment were preempted by the Copyright Act. Thus, the Court declined to reach the preemption issue in its Order.

The author plaintiffs are expected to file an amended complaint to try to address the issues raised by the Court in granting the OpenAI defendants’ initial motions to dismiss.  Notably, the Court has consolidated the two cases given that they involve the same claims, factual allegations and putative class members.  If an amended complaint is filed, it is likely the OpenAI defendants will again move to dismiss the claims and may raise the preemption issue as to the state law claims.  In the meantime, similar lawsuits are being filed against other AI companies.