A proposed bill in California that would force generative artificial intelligence (AI) developers to inform copyright owners when their materials are included in training datasets is the latest stab by lawmakers to protect content creators in the age of GenAI.
“As the AI industry continues to develop and expand, it is critical for content creators to know if and how their work is being used to train advanced models,” State Assemblymember, Rebecca Bauer-Kahan (D-Orinda), author of AB 412, the “AI Copyright Transparency Act,” said in a statement. “The AI Copyright Transparency Act increases accountability for AI developers and empowers copyright owners to exercise their rights.”
The bill – sponsored by the Screen Actors Guild-American Federation of Television and Radio Artists (SAG-AFTRA) and co-sponsored by the Concept Art Association” (CAA) and the National Association of Voice Actors (NAVA) – is aimed at holding accountable developers who use copyrighted materials to train new GenAI models without crediting or compensating the materials’ owners.
In absence of any significant tech legislation this century, state lawmakers – particularly in California – have assumed the mantle of leading the way. Recent laws in the Golden State have tackled privacy rights (California Consumer Privacy Act), security (The Internet of Things Security Law), autonomous vehicles, and AI safety, especially around deepfakes.
But the issue surrounding AI, original content and training models is a conundrum that has bedeviled the entertainment industry in particular and been a major focus of tension in union conflicts between creators and the studios they often work for. In fact, Hollywood has been as fraught with tension about AI as any industry in terms of content creation, dating to Baidu Inc.’s DeepSpeech technology that first converted speech into text a decade ago.
“Copyright and ownership of sound files has become an issue at the forefront of our industry, and the lack of legislation around these new technological systems has negatively and tangibly impacted all entertainers,” said The National Association of Voice Actors, referring to loss of jobs and projects by its members to AI-replicated voices.
Yet there is no national AI legislation of any sort, and the courts have yet to weigh in on the legality of the practice, wherever content is made.
A handful of federal bills have circulated in Congress in recent years, but federal lawmakers have had next to no success on tech legislation for nearly a quarter-century.
A Senate bill introduced in November by Sen. Peter Welch, D-Vt., would in his words “set a higher standard for transparency.” The Transparency and Responsibility for Artificial Intelligence Networks (TRAIN) Act would enable copyright holders to subpoena training records of generative AI models, if the holder can declare a “good faith belief” that their work was used to train the model.
Nearly a year ago, then-Rep. Adam Schiff, who is now a California Senator, introduced the Generative AI Copyright Disclosure Act that would require AI companies to outline every copyrighted work in their datasets.
