Tremblay v. OpenAI, Inc.

Background

This proposed class action concerns authors’ allegations that OpenAI copied their copyrighted books and used them as training data for language models that operate ChatGPT. The plaintiffs hold registered copyrights in their books and allege that ChatGPT can generate accurate summaries of those books. They asserted claims for direct copyright infringement, vicarious copyright infringement, violation of Section 1202(b) of the Digital Millennium Copyright Act, unfair competition under California law, negligence, and unjust enrichment.

OpenAI’s motions to dismiss sought dismissal of every claim except direct copyright infringement. The court evaluated the challenged claims under Federal Rule of Civil Procedure 12(b)(6), which permits dismissal when a complaint does not state a legally sufficient claim. At this stage, the court treated factual allegations as true and viewed them in the plaintiffs’ favor, but it did not accept conclusory allegations without supporting facts.

Rulings

Vicarious copyright infringement

The court dismissed Count II with leave to amend. Vicarious infringement requires an underlying direct infringement and allegations that the defendant had the right and ability to supervise the infringing conduct and a direct financial interest in it. The court did not reach OpenAI’s latter two arguments because it found that the plaintiffs had not adequately alleged the required direct infringement. The plaintiffs alleged that every model output was an infringing derivative work, but they did not identify any particular output or allege that an output was substantially similar to their books. The court therefore concluded that the pleadings did not adequately allege direct copying or substantial similarity.

Digital Millennium Copyright Act

The court dismissed the Section 1202(b) claims with leave to amend. That provision addresses the intentional removal or alteration of copyright-management information, such as an author’s name or a work’s title, and certain distribution of works or copies knowing that the information was removed or altered.

For the Section 1202(b)(1) theory, the plaintiffs alleged that OpenAI designed its training process to remove copyright-management information. The court found those allegations conclusory and noted that examples of ChatGPT outputs referred to the plaintiffs’ names. It also found that the plaintiffs did not adequately allege facts showing that OpenAI knew, or had reasonable grounds to know, that removing the information during training would induce, enable, facilitate, or conceal infringement.

For the Section 1202(b)(3) theory, the court found that the plaintiffs did not allege that OpenAI distributed their books or copies of their books. Alleging that ChatGPT outputs were derivative works, without describing those outputs or showing that they were the books or copies of the books, was insufficient.

Unfair competition

The plaintiffs asserted California unfair-competition claims under the unlawful, fraudulent, and unfair theories. The court rejected the unlawful theory because it depended on the dismissed Digital Millennium Copyright Act claims and because the complaints did not allege that OpenAI reproduced and distributed copies of the books. The court rejected the fraudulent theory because the plaintiffs did not identify adequate fraud allegations and did not satisfy the heightened pleading requirements for that theory.

The court allowed the unfair theory to proceed. Accepting the allegations for purposes of the motions, the court concluded that using the plaintiffs’ copyrighted works to train language models for commercial profit may constitute an unfair business practice.

Negligence

The court dismissed the negligence claim with leave to amend. The plaintiffs alleged that OpenAI negligently collected, maintained, and controlled systems trained on their copyrighted works. The court found that the plaintiffs did not identify a legal duty requiring OpenAI to maintain and control the public information contained in their books. It also rejected the argument that the parties had a special relationship, noting that the complaints did not allege a fiduciary or custodial relationship.

Unjust enrichment

The court found that the unjust-enrichment claim failed. Under the theory presented, the plaintiffs needed to allege that OpenAI received and unjustly retained a benefit at their expense and that the benefit was conferred through mistake, fraud, coercion, or request. The court found that the plaintiffs had not made those allegations. The opinion does not separately state in its conclusion whether this claim’s dismissal was with or without leave to amend.

Disposition and case management

Judge Araceli Martinez-Olguin granted in part and denied in part the motions to dismiss. The amended complaint was due March 13, 2024, and the court directed that it consolidate claims in the Tremblay case, the Silverman case, and a third related case. The court stated that no additional parties or claims could be added without the court’s permission or the defendants’ agreement, and it said it would issue a separate order consolidating the cases.