Summary

Visual artists sued Google last week, alleging that Google’s AI-powered image generator, Imagen, was trained on their copyrighted content without authorization. The proposed class action asserts claims of direct copyright infringement against Google and vicarious infringement against Google’s parent company, Alphabet. The artists’ suit relies on the same theory underlying the string of recent litigation; the unauthorized inclusion of copyrighted content in the underlying datasets used to train large-scale AI models is copyright infringement.

The Upshot

• A new AI copyright infringement case dropped this week, this time against Google and its parent company, Alphabet. Four visual artists claim Google’s AI-powered image generator, Imagen, used their copyrighted art without approval.
• The plaintiffs are seeking an unspecified amount of monetary damages, and an order forcing Google to destroy all copies of their work.
• This latest case is yet another iteration of a pattern emerging in suits against AI companies; that the inclusion of creators’ content in training sets for AI models constitutes direct copyright infringement.

The Bottom Line

With many AI-copyright-infringement cases currently underway, the legal future of AI remains murky. It will take time for the courts to substantively assess the merits of the copyright claims, as well as the defendants’ expected defenses, such as fair use.

Several visual artists are throwing their hats into the AI-copyright-infringement ring this week, adding another suit to the growing number of copyright litigations against Artificial Intelligence (AI) companies. The group of artists sued Google in the U.S. District Court for the Northern District of California on April 26, 2024, for copyright infringement based on Google’s Imagen image generator, alleging Google trained it on their copyrighted content without their authorization.

There are four named plaintiffs in the suit, Zhang v Google LLC and Alphabet Inc, No 3:24-CV-02531: Jinga Zhang, a photographer from Washington; and Sarah Andersen, Hope Larson, and Jessica Fink, cartoonists and illustrators from Oregon, North Carolina, and New York, respectively. The four plaintiffs have proposed the suit as a class action on behalf of all U.S. persons and entities that own a U.S. copyright in a work that Google used as training image(s) for the image generator. Ms. Andersen is no stranger to AI-related copyright litigation. She is a named plaintiff in one of the first suits of this nature, a similar action against Stability A.I., also in the Northern District of California, based on image generators Stable Diffusion and Midjourney. Her claim of direct copyright infringement against that particular company survived a motion to dismiss last fall. Ms. Zhang has been added as a named plaintiff in that case as well.

Imagen, the AI tool at issue in this newest suit, is a text-to-image diffusion model. Users input a text description into the tool, which then uses AI techniques to generate a responsive image. Plaintiffs allege that in 2022, Google publicly announced Imagen’s existence and stated that it was trained on the publicly-available LAION-400M dataset, a list of 400 million image-text pairs. Each image is paired with text that includes a URL that links to the image. Plaintiffs allege this dataset contained their copyrighted works. The Plaintiffs describe Google’s gradual rollout of Imagen to increasingly large numbers of consumers, culminating in the December 2023 release of Imagen 2, the successor to Imagen. The Complaint contains several detailed paragraphs about the development of the LAION-400M dataset and how its implementation requires a user to acquire copies of the actual training images from URLs, thereby proving that Google copied Plaintiffs’ copyrighted images without authorization by virtue of utilizing the LAION-400M dataset. In addition to a direct copyright infringement claim against Google, the complaint asserts a vicarious copyright infringement claim against Alphabet, Google’s parent company. Plaintiffs demand, among other relief, statutory damages, and “destruction or other reasonable disposition of all copies Defendants made or used in violation” of Plaintiffs’ exclusive rights.

This latest case is yet another iteration of a pattern emerging in suits against AI companies; that the inclusion of creators’ content in training sets for AI models constitutes direct copyright infringement. However, plaintiffs here appear to have taken note of developments in earlier suits, particularly the string of successful motions to dismiss where AI companies have succeeded in at least narrowing, and in some instances, dismissing, the claims against them, and tailored their complaint accordingly. For example, in Ms. Andersen’s earlier litigation against Stability A.I., other plaintiffs were dismissed from the suit because they did not hold copyright registrations (which are a prerequisite to bringing suit). Here, the named Plaintiffs attached exhibits of their copyright registrations they assert against Defendants. Likewise, the Defendants in Ms. Andersen’s earlier suit argued that plaintiffs need to identify which specific works were copied by their image model, and here, Exhibit B contains a “non-exhaustive list of copyrighted images registered by Plaintiffs and infringed by Defendants,” purportedly included in LAION-400M, the training dataset for Google’s Imagen.

As with the many other copyright suits brought against AI companies under this theory of unauthorized inclusion of content in training datasets, it will take time for the courts to substantively assess the merits of plaintiffs’ copyright claims and defendants’ expected defenses like fair use.

[View source.]