Two bipartisan senators introduced a bill Tuesday that would force AI companies to disclose copyrighted material contained in datasets used to train AI models. of Copyright Notice and Ethical AI Reporting (CLEAR) Act Before making a model available to the public, developers must submit a notice to the Copyright Registry detailing the work used in training. This will also apply retroactively to models that have already been released. Registrants must maintain a public database of submitted notices.
“AI has the potential to improve our lives and change the way we work and innovate, but we need a unified approach to put in place guardrails that protect the jobs and lives of all workers, including artists and creators,” said Sen. Adam Schiff (D-Calif.), one of the bill’s two sponsors. in a statement. “I am proud to join Senator Curtis in introducing this bipartisan bill. It is an important step toward ensuring transparency in the development of AI models and ensuring creators are fairly compensated for their work.”
Co-sponsor Sen. John Curtis (R-Utah) added, “Congress must help accelerate AI innovation, but not without transparency and accountability.” “By shining a light on how generative AI models are trained, our bipartisan bill will help build public trust in emerging technologies and foster America’s best creativity.”
The bill is not Schiff’s first push for AI data transparency. While still in the House of Commons in 2024, he introduced: Generative AI Copyright Disclosure Lawcovers most of the same areas. Like previous bills, the new law would impose civil monetary penalties for failure to disclose the copyrighted material used.
Related: New York bill would require disclosure of AI-generated material in news articles
Data transparency has been a key demand of creators and the copyright industry since the early days of the generative AI boom, but AI developers have resisted the cost and complexity of identifying tens of millions of works in training datasets, as well as concerns about triggering lawsuits from rights holders.
Transparency measures are included in the European Union’s AI law, but disclosure requirements have never garnered enough support to become law in the U.S. Congress. California’s Training Data Transparency Act requires developers to provide summaries of the datasets used in training, but the summaries released so far are lack of details.
The CLEAR Act also stops short of requiring a license for copyrighted works used in training. This is another key demand of creators and is also the subject of multiple ongoing lawsuits against developers by authors and rights holders.
Nevertheless, the bill was quickly approved by an array of creative industry groups, including the Writers Guild, Writers Guild of America, National Association of Music Publishers, Recording Industry Association of America, and SAG-AFTRA. Notably, however, the Motion Picture Association (MPA), which represents major Hollywood studios, was missing from the list and had to overcome disagreements among its members over how aggressively to pursue legal action.
Also not on the list is the Alliance of Motion Picture and Television Producers (AMPTP), which represents film and television production companies. AMPTP currently represents film and television production companies; New round of labor-management negotiations Negotiations with the Actors and Writers Guild, where the use of AI is expected to play a major role in the negotiations. The industry suffered a months-long production shutdown in 2023 after last-minute contract negotiations collapsed, largely over the role of AI. Publicly supporting a guild-backed AI bill now could weaken producers’ leverage in future negotiations.
