In a prior post, I summarized four United States copyright decisions that examined copyright infringement liability for training and making available generative AI (GenAI) services. It focused, among other things, on whether:
- downloading and copying of works (or other copyright subject matter) to train GenAI models infringes copyright;
- the trained GenAI models are infringing because they contain or embody copies of training materials in a material object or material form or are, under US law, derivative works;
- all outputs from GenAI systems are derivative works because they are derived from copyright works and whether specific outputs can infringe the copyrights in specific training materials; and
- fair use is potentially available in the U.S. as a defense to GenAI copyright infringement claims.
Since then, U.S. District Courts have released two more important decisions: Doe v. GITHUB, INC., (N.D. Cal. Jan. 3, 2024) (“Github”) and Tremblay v. OpenAI, INC., (N.D. Cal. Feb 12, 2024) (“Tremblay”).
These decisions are slowly witling away the plaintiffs’ claims under the DMCA which prohibits the intentional removal or alteration of copyright management information (CMI) or the distribution of works knowing that CMI has been removed or altered without the authority of the copyright owner. They are similarly witling away at the plaintiffs’ California state law claims.
Github State Law Claims
In Github, a group of software developers sued OpenAI and GitHub for allegedly violating their copyrights and open-source licenses by using their code to train and operate Copilot, an AI system that generates code suggestions based on natural language inputs. The plaintiffs claimed that OpenAI and GitHub removed or altered the CMI from their code and distributed it without proper attribution or license terms, in breach of the Digital Millennium Copyright Act (DMCA) and various contracts. The defendants did not move to dismiss the plaintiffs’ claims for breach of contract for open-source license violations or breach of contract for selling licensed materials. They did, however, move to dismiss the asserted state law claims for interference, fraud, unfair competition, and negligence.
The Court dismissed the plaintiffs’ state law claims for intentional and negligent interference with prospective economic relations, unjust enrichment, unfair competition, and negligence as being preempted by Section 301 of the U.S. Copyright Act.
The Ninth Circuit has established a two-part test to determine whether state law claims are preempted by the copyright law. As summarized by the Court, first, the court decides whether the subject matter of the state law claim falls within the subject matter of copyright. If it does, the court must then determine whether the rights asserted under state law are equivalent to the rights contained in the copyright law. On the other hand, if a state law claim includes an ‘extra element’ that makes the right asserted qualitatively different from those protected under the U.S. Copyright Act, the state law is not preempted by the Copyright Act.
The court held that these state law claims were essentially based on the same rights that were protected by the U.S. Copyright Act, such as the rights of reproduction, distribution, and preparation of derivative works, and that they did not include any extra element that would make them qualitatively different from those rights. The court dismissed these claims with prejudice, meaning that the plaintiffs could not reassert them in a revised complaint.
Github DMCA Claims
The Defendants sought to dismissed the CMI claims under the DMCA arguing that they only apply when CMI is removed or altered from an identical copy of a copyrighted work. The court agreed and dismissed the claims, with leave to amend.
Under the DMCA, CMI includes information such as the title, the author, the copyright owner, the terms and conditions for use of the work, and other identifying information set forth in a copyright notice or conveyed in connection with the work. To state a claim under Section 1202(b)(1) in the Ninth Circuit, Plaintiffs must plausibly allege that Defendants (1) intentionally removed or altered CMI while knowing, or having reasonable grounds to know, that it will induce, enable, facilitate, or conceal an infringement of any right under the DMCA. Similarly, a violation of Section 1202(b)(3) requires that Plaintiffs plausibly allege that Defendants distribute or import for distribution copies of works knowing that CMI has been removed or altered without authority of the copyright owner.
The Court agreed with the Defendants that no DMCA violation exists where the allegedly infringed and infringing works are not identical. The Court rejected the DMCA violations related to CMI as the alleged copyright infringements were that the output from Copilot “is often a verbatim copy, even more often it is a modification: for instance, a near-identical copy that contains only semantically insignificant variations of the original Licensed Materials, or a modified copy that recreates the same algorithm.” Some examples provided by the Plaintiffs alleged that the Copilot output is a “modified format,” “variation[],” or the “functional[] equivalent” of the licensed code.
Tremblay State Law Claims
Like in the Github case, OpenAI moved to dismiss certain state law claims, here, for Unfair Competition under Cal. Bus. & Prof. Code § 17200 (UCL); for negligence; and for unjust enrichment.
The Court dismissed the plaintiffs’ claims for unfair competition, negligence and unjust enrichment under California law for the following reasons.
Unfair Competition: The plaintiffs alleged that the defendants engaged in “unlawful business practices” by violating the DMCA. The Court held that the plaintiffs failed to allege an economic injury caused by the defendants’ alleged removal of CMI. The plaintiffs argued that they lost intellectual property in connection with the DMCA claims because of the “risk of future damage to intellectual property that results the moment a defendant removes CMI from digital copies of Plaintiffs’ work — copies that can be reproduced and distributed online at near zero marginal cost.” However, nowhere in plaintiffs’ complaint did they allege that defendants reproduced and distributed copies of their books. Accordingly, any injury was considered as speculative.
The Court also dismissed the UCL claim based on fraudulent conduct, finding that the plaintiffs failed to allege facts showing that the defendants had the required mental state to remove CMI knowing that it would induce or conceal infringement. The Court further found that the plaintiffs failed to plead allegations of fraud with particularity as required.
The Court allowed the UCL claim based on unfair conduct to proceed, finding that the plaintiffs adequately alleged that the defendants’ use of the plaintiffs’ copyrighted works to train ChatGPT for commercial profit may constitute an unfair practice.
Negligence: Under California law, negligence claims require that a plaintiff establish (1) duty; (2) breach; (3) causation; and (4) damages. The Court held that the plaintiffs failed to allege that the defendants owed them a legal duty to safeguard their works. The Court rejected the plaintiffs’ argument that there was a special relationship between the parties, finding no fiduciary or custodial relationship. The Court also noted the possibility that the economic loss doctrine may bar recovery for negligence. The Court dismissed the negligence claim with leave to amend.
Unjust Enrichment: As summarized by the Court, under Ninth Circuit law, an unjust enrichment claim can be advanced as either an independent cause of action or as a quasi-contract claim for restitution. To allege unjust enrichment as an independent cause of action, a plaintiff must show that the defendant received and unjustly retained a benefit at the plaintiff’s expense. Restitution is not ordinarily available to a plaintiff unless the benefits were conferred by mistake, fraud, coercion or request; otherwise, though there is enrichment, it is not unjust. The theory underlying an unjust enrichment claim is that a defendant has been unjustly conferred a benefit `through mistake, fraud, coercion, or request.
The Court held that the plaintiffs failed to allege that the defendants unjustly obtained benefits from the plaintiffs’ copyrighted works through fraud, mistake, coercion, or request. The Court dismissed the unjust enrichment claim with leave to amend.
Unlike in the Github case, the Court did not decide the state law claims based on the copyright pre-emption doctrine. However, the Court noted in footnotes that the UCL claim “may be preempted by the Copyright Act” and that because “Plaintiffs fail to adequately allege the negligence and unjust enrichment claims, the Court need not reach the preemption issue at this time.”
Tremblay DMCA Claims
The court dismissed the plaintiffs’ claims that OpenAI violated Section 1202(b) of the DMCA.
As noted above, to state a claim under Section 1202(b)(1), the plaintiffs had to show that OpenAI intentionally removed or altered CMI from the books used to train ChatGPT, and that it did so with the knowledge or reasonable grounds to know that it would induce or conceal infringement. The court held that the plaintiffs failed to allege any facts to support this claim, such as how OpenAI removed or altered CMI, what CMI was removed or altered, and how that removal or alteration facilitated or concealed infringement.
As also noted above, to state a claim under Section 1202(b)(3), the plaintiffs had to show that OpenAI distributed or imported for distribution the plaintiffs’ books or copies of their books, knowing that CMI had been removed or altered without authority. The Court held that the plaintiffs failed to allege this as well, because the ChatGPT outputs were not identical to the plaintiffs’ books and therefore did not constitute copies of the books.
The court gave the plaintiffs leave to amend their DMCA claims.
Github and Tremblay Copyright Infringement Holdings
Both cases followed the prior decisions summarized in my prior blog.
In Github, the court dismissed a motion to dismiss standing claims that Copilot output could not infringe copyright where specific instances in which their code was output by Copilot was pleaded.
In Tremblay, the Court again rejected the vicarious copyright infringement claim that was premised on the direct infringement theory that every “every output of the OpenAI Language Models is an infringing derivative work”. The Court re-iterated that the authors failed to allege that OpenAI’s language model outputs were substantially similar to their books, which is a necessary element for a vicarious copyright infringement claim. The Court, once again, dismissed these claims with leave to amend.
Comments on Github and Tremblay Decisions
These two recent decisions signal the likelihood that the copyright legality of GenAI systems training will be resolved in the U.S. under that country’s copyright law. If the decisions referred to in my prior blog are any indication, the most likely issue is whether the fair use defense will be applicable.
The decision in the Github case applying the U.S. pre-emption doctrine raises an interesting question as to whether a similar result might be arrived at in Canada. The Canadian Copyright Act does not have a provision that is a direct analogue of Section 301 of the U.S. Act. However, Section 89 of the Copyright Act provides that:
No person is entitled to copyright otherwise than under and in accordance with this Act or any other Act of Parliament, but nothing in this section shall be construed as abrogating any right or jurisdiction in respect of a breach of trust or confidence.
This provision was applied by the Supreme Court of Canada in the value for signal case to strike down a CRTC regulatory regime that the Court found would have created a functional equivalent right to a copyright that Parliament had not sought to prohibit under the Act. The Court also explained the two types of paramountcy the Court has recognized to address conflicts between federal and other laws: operational conflict (which arises when there is an impossibility of compliance with both provisions) and incompatibility of purpose (the conflict arises because applying one provision would frustrate the purpose intended by Parliament in another).