Get Full Government Meeting Transcripts, Videos, & Alerts Forever!
Get email alerts on the Technology Policy topic
No spam. Unsubscribe anytime.
Assembly approves disclosure requirement for AI model training data after hours of testimony from creators and tech groups
Summary
AB 412 would let copyright holders request whether a generative AI model was trained on their works; creators backed the idea as a minimal transparency right, while tech and industry groups warned of trade-secret and preemption issues.
The Assembly Judiciary Committee on Tuesday approved AB 412, a bill that would give creators a right to ask developers whether a generative AI model was trained on specific copyrighted works.
Sponsor Assemblymember Laura Bauer-Kahan said the bill does not change federal copyright law but provides a narrowly targeted transparency right so that creators — from voice actors to visual artists and musicians — can learn whether their work was used to train a model. “How does someone who owns a copyright ... exercise their rights in today's world where there are models that are being trained on those copyrighted works?” Bauer-Kahan asked the committee.
Voice and other creative professionals packed the hearing in support. Tim Friedlander, president of the National Association of Voice Actors, told the committee, “AI companies are profiting off of our life's work, and there's currently no law in place to stop them or even for us to find out if our work has been used to train their models.” Several artists and guild representatives testified the bill would be a practical first step to allow rights holders to determine whether they should pursue licensing or legal action.
Opponents included the California Chamber of Commerce, the Recording Industry Association of America and technology trade groups. They argued the bill overlapped with federal copyright law and ongoing litigation, could be preempted, and risked exposing proprietary training pipelines. The Chamber urged delay until the U.S. Copyright Office issues forthcoming guidance and warned of trade-secret claims and litigation risks.
Assembly staff and several members noted the bill aims to provide a lower-cost, administrative path for rights holders who otherwise cannot get that information without burdensome discovery. Attorney Travis Manfredi, who has litigated AI copyright cases, said courts often dismiss claims when plaintiffs cannot show use of a work and that the proposed disclosure mechanism fills that gap.
Committee members asked whether academic or noncommercial models should be exempted. The author accepted an amendment exempting models developed and used exclusively for noncommercial academic or government research; industry groups said the carve-out must be tightly drawn to prevent misuse. The committee passed the bill as amended and sent it forward.
