In the wake of several Congressional hearings over the past year on AI and intellectual property, Representative Adam Schiff (D-California) has introduced the Generative AI Copyright Disclosure Act of 2024 (H.R. 7913). The proposed law addresses concerns over lack of transparency in the data sets used to train generative AI models by requiring submission of a notice to the United States Copyright Office regarding all copyright-protected works used to train a particular AI system prior to the release of that system. Although the bill has almost no chance of gaining traction in Congress, it showcases concerns that current U.S. copyright law may not adequately address new generative AI technologies.
Act Particulars
The Act’s notice requirement would apply to those who create or significantly alter “a training dataset,” which the bill defines as “a collection of individual units of material (including a combination of text, images, audio, or other categories of expressive material, as well as annotations describing the material) used to train a generative AI model” and build a generative AI system. The notice would need to include a “sufficiently detailed summary” of any and all copyrighted works—whether registered or not—used in the training dataset or to alter the training dataset. Moreover, if the dataset is publicly-available when the notice is submitted, the notice would need to include the URL where the dataset appears. If a required notice is not submitted pursuant to certain deadlines set forth in the Act, violators would be a subject to a civil penalty of at least $5,000. The Act would take effect within 180 days of enactment, and would require the U.S. Copyright Office to issue implementing regulations within that same timeframe. The Copyright Office would also need to establish a publicly-accessible online database consisting of each notice filed.
Stakeholders’ Perceptions
The interests of those creating or altering the generative AI datasets are no doubt at odds with those of copyright owners. We predict that, generally, copyright owners’ perceptions of the Act will be very positive, while generative AI companies will have a very negative reaction. This is not surprising given the challenges generative AI companies would face in tracking down individual rights holders, and then, once the rights holders are on notice of what is going on, likely having to obtain formal licenses and pay them royalties before incorporating their works in training datasets. Such requirements could be viewed as chilling technological growth and promoting AI monopolies, since, for example, smaller or niche companies would likely be unable to pay the requisite costs. The effects could also have a negative spillover effect on rightsholders because any royalties paid to them would be very low even for financially successful AI models, given that the royalties would need to be spread among millions of creators. Moreover, AI companies may focus on compensating larger copyright holders, not only because they own the rights to a large volume of works, but also because their rights are likely more well-known or easily discovered by generative AI creators or editors. As a result, individual creators’ rights may be overlooked, even if generative AI companies make good faith efforts to identify and pay them for use of their protected works.
On the other hand, many rightsholders currently feel as though AI companies are unfairly profiting from their works without permission and without compensating the rightsholders, despite copyright protections being enshrined in the U.S. Constitution. As early decisions in AI-related copyright litigation indicate (see our prior coverage here), copyright owners may have a difficult time preventing the unauthorized use of their works to train AI models under the current version of the U.S. Copyright Act. Accordingly, many content creators and rights owners seem to be embracing the bill as a step in the right direction to avoid unfair infringement and provide protection that may be otherwise difficult to achieve. From the perspective of copyright owners, protecting their works encourages the creation of new works—after all, if you create a work, and then someone can copy it and profit it from it, without paying you, why spend the time to create a new work? This, they argue, is the reason that copyright protection exists in the first place.
What Does the Future Hold?
While strong opinions have been voiced on both sides of this issue, it bears noting that the Act has no co-sponsors to date and is unlikely to receive much consideration in Congress. Even so, the bill reflects the fact that there is significant public support, particularly among writers, artists, and other creatives, for copyright protections that address the advent of generative AI. But it also highlights the thorny issues that come with attempting to regulate new technologies, including the possibility that regulations could inadvertently stifle new technological development. This is a debate that will continue not only in Washington, D.C., but throughout the country.
In the meantime, it remains to be seen how courts addressing AI-related issues will decide. Those cases may provide guideposts for Congress to further draft and propose an AI act that would better suit the involved parties.
