EN | 中文

Navigating Copyright in AI: The Generative AI Disclosure Act of 2024 and Its Implications, By Aimee Scala, Partner and Hanoch Sheps, Counsel

On April 9, 2024 a bill titled the “Generative AI Disclosure Act of 2024”  (https://www.congress.gov/bill/118th-congress/house-bill/7913/text) was introduced in the U.S. House of Representatives concerning the use of copyrighted works in building generative AI systems.

If enacted as drafted, the law would require anyone who creates (or significantly modifies) a training dataset which contains copyrighted works in connection with the development of a generative AI system to submit a notice to the Register of Copyrights containing “a sufficiently detailed summary” of the copyrighted works used in the training dataset as well as a link to a URL containing the dataset. Failure to submit the required notice would result in a civil penalty of at least $5,000. The law would also require the Copyright Office to maintain a public database of the notices submitted on its website and require the Register of Copyrights to issue regulations to regulate the required disclosures within 180 days of the effective date of the act.

The disclosure requirements and other provisions would become effective 180 days after the bill is signed into law. At that point, generative AI providers who have trained their systems on copyrighted works and who have already released their systems to the public would have 30 days to file the required notice. For those generative AI systems using copyrighted works in the training dataset released to the public after the effective date, those providers would have to file the required notice at least 30 days before public release of the system.

Coming almost a year after the subcommittee on Courts, Intellectual Property, and the Internet of the House of Representatives conducted hearings on the impact of artificial intelligence on intellectual property, these disclosure requirements would provide more transparency to the public—including those authors whose works are being used in AI training datasets without their knowledge—and appear to be a step towards addressing the copyright issues that arise in connection with the so-called “input” side of generative AI systems.

With respect to the “output” side of generative AI systems, i.e., what the system produces in response to a user’s prompt, the Copyright Office has already taken the position that these works are not entitled to copyright protections because they lack human creativity. Accordingly, a comic book created using the generative AI system Midjourney was not entitled to copyright protection (https://www.copyright.gov/docs/zarya-of-the-dawn.pdf); neither was a two-dimensional artwork created using Midjourney that won the 2022 Colorado State Fair’s annual fine art competition (https://www.copyright.gov/rulings-filings/review-board/docs/Theatre-Dopera-Spatial.pdf). This human creativity requirement means that works created by, by and large, the outputs of AI generative systems cannot be protected by copyright.

This position, while disincentivizing the replacement of human creators such as writers, illustrators, artists, and others in various creative fields, does not appear to be the position taken by other nations. For example, English law employs the Copyright, Designs and Patents Act of 1988 allowing creative works made by computers to enjoy copyright protections. (https://www.legislation.gov.uk/ukpga/1988/48/section/178). While Canadian law is less clear, it has similarly given copyright protection to works created with the assistance of AI if it involves human skill and judgment from the human co-author. It remains to be seen whether U.S. creative industries will see creators begin to work in other jurisdictions where works created using generative AI tools can be protected, and whether Congress might step in to provide legislation on the “output” side of the AI generative systems equation.

This bill was introduced against the backdrop of at least 24 pending copyright lawsuits against AI companies in the U.S. and over 1,000 AI policy initiatives in 69 countries, according to the AI Policy Observatory live tracker of the Organisation for Economic Co-operation and Development (OECD). In addition, at least 20 U.S. States have proposed legislation on the topic, and there has been steps taken in the private sector to put new corporate policies and contractual provisions in place to address the proliferation of AI technologies.

One such litigation, Andersen v. Stability AI Ltd. (N.D. Cal.) seeks to address claims of infringement for both the “input” and “output” side of the generative AI systems Midjourney, Stability, DeviantArt, and Runway. On the “input” side of the equation, the parties are addressing whether using copyrighted works as training data for input into AI image generation platforms constitutes broad infringement of the reproduction rights in those works, or if instead, training generative systems using copyrighted works is instead fair use given the arguably transformative nature of the systems. The training datasets involved are believed to contain 400 million training images and 5.85 billion training images, respectively. On the “output” side, issues concerning infringement arise with respect to the way these systems mimic artists’ styles – intentionally creating images that resemble genre and style raising issues concerning derivative copying and licensing. However, because system outputs often generate new images based on, in some cases billions of training images, it is unclear whether the AI-generated images can be considered substantially similar for infringement purposes, whether any use of any particular copyrighted work is de minimis, and if they are substantially similar, whether the resulting works are nevertheless fair use of the copyrighted works.

There is a pending motion to dismiss the putative class action amended complaint with far-reaching implications with respect to the development of judicial doctrine addressing these questions. For now, it remains to be seen what additional compliance and regulatory implications the Generative AI Disclosure Act of 2024 and other proposed legislation will have on the marketplace and what impact the pending lawsuits will have on the legislative and regulatory process.