In a move aimed at addressing concerns over copyright infringement in the field of artificial intelligence (AI), U.S. Representative Adam Schiff (D-Calif.) has introduced the Generative AI Copyright Disclosure Act in the House of Representatives. The bill, if passed, would compel AI companies to disclose the copyrighted works used to train their models or face financial penalties.

Under the proposed legislation, both new and previously released generative AI systems would be subject to the requirement. Companies would be mandated to file a comprehensive list of copyrighted works included in an AI model's training data set with the Copyright Office at least 30 days before making the model available to consumers. In addition, any significant alterations to an existing model's training data set would necessitate a new disclosure.

The financial penalties for non-compliance would be determined on a case-by-case basis by the Copyright Office, taking into consideration factors such as the company's past history of noncompliance and its size.

Generative AI models are trained on vast amounts of existing works, ranging from film scripts to news articles to music. While some data sets are acquired through licensing agreements with copyright owners, many AI models scrape the internet for content, including copyrighted material, without the knowledge or consent of the authors. This practice, often defended by large AI companies as "fair use," has ignited a debate and led to numerous lawsuits between copyright owners and AI companies.

One such lawsuit was filed last October by music publishers including Universal Music Group, ABKCO, and Concord Music Group against AI giant Anthropic for allegedly unlawfully exploiting their copyrighted song lyrics to train AI models. The publishers accused Anthropic of copying and disseminating copyrighted works on a widespread basis.

While some in the creative industries are seeking compensation and the ability to opt in or out of being included in AI data sets, Schiff's bill specifically focuses on transparency regarding copyrighted training data. Nevertheless, the legislation has garnered support from various music industry groups such as the Recording Industry Association of America (RIAA), the National Music Publishers' Association (NMPA), and the Black Music Action Coalition (BMAC), among others. It has also received backing from creative industry groups including the Professional Photographers of America, SAG-AFTRA, and the Writers Guild of America.

Rep. Schiff emphasized the importance of balancing the potential of AI with ethical guidelines and safeguards for creators, stating that the Generative AI Copyright Disclosure Act champions innovation while ensuring creators are aware of how their work contributes to AI training datasets.

The introduction of the bill has been met with approval from rights groups as well. RIAA Chief Legal Officer Ken Doroshow praised Schiff for addressing the issue of comprehensive and transparent record-keeping as a critical element in AI regulation. NMPA President/CEO David Israelite emphasized the need for AI companies to reveal the works used to train their data in order to fully license and fairly compensate songwriters. ASCAP CEO Elizabeth Matthews echoed this sentiment, stressing the importance of placing humans first and protecting the rights of creators.

With the Generative AI Copyright Disclosure Act, Schiff intends to foster responsible AI development that respects and values the contributions of creators while establishing clear rules and transparency for the use of copyrighted materials in AI systems.