Support AB 412: Generative Artificial Intelligence: training data: copyrighted materials
July 8, 2025
Honorable Tom Umberg
Chair, Senate Judiciary Committee
1021 O Street, Suite 3240
Sacramento, CA 95814
Subject: Support AB 412: Generative Artificial Intelligence: training data: copyrighted materials
Dear Chair Umberg,
The Digital Media Licensing Association (DMLA) is a non-profit, non-partisan trade association founded in 1951 that represents the interests of content creators, digital media producers, distributors, and licensors. Our membership comprises thousands of industry professionals across the visual content ecosystem, ranging from individual photographers and illustrators to major image licensing companies, news wire agencies, technology innovators, and AI developers. For more than seven decades, we have worked to establish business standards, develop best practices, and advocate for copyright protection, privacy rights, fair licensing practices, and now ethical AI development. DMLA members license hundreds of millions of images, videos, illustrations, vectors, audio, and other creative content globally each day, powering everything from news media and educational materials to corporate communications, advertising campaigns, and entertainment products.
AB 412, drafted by Assembly Member Rebecca Bauer-Kahan, increases transparency around the use of copyrighted materials to train generative artificial intelligence by requiring GenAI developers to provide copyright owners with information about how their materials are used.
GenAI developers frequently use copyrighted materials to train new systems without crediting or compensating the lawful owners of those materials. Federal law provides copyright holders with exclusive rights to reproduce, distribute, and display their works. To exercise these rights, owners must first know when and how their works are being used.Currently, copyright owners have no way of knowing whether their copyrighted materials have been used to train a given GenAI model.
This bill requires a developer of a GenAI system or model to, upon receiving a request from a copyright owner, provide that owner with a list of the copyrighted materials held by the owner that were used to train the system or model. The bill further authorizes a copyright owner to bring a civil action against a developer who fails to respond to a properly submitted request.
Given our technical subject-matter expertise, DMLA also wishes to address two specific issues opponents have raised:
a) Fingerprinting feasibility. Opponents question whether fingerprinting can reliably match and identify copyrighted content in training data. Fingerprinting technology in the image industry has been available for over twenty years, and DMLA members have included a number of image recognition technology companies going back to 2002 with Picscout, one of the first in the market. Current DMLA member ImageRights International, Inc. (www.imagerights.com) has employed proprietary image fingerprinting technology for well over a decade. ImageRights generates a unique fingerprint for each copyrighted image a copyright owner uploads to its system, then continuously scans websites, social media platforms, and online marketplaces. ImageRights fingerprints every image it discovers and compares them to all of the fingerprints in its database of copyrighted works. When a match is found, ImageRights automatically notifies the lawful copyright owner. A similar system could be implemented to allow creators to determine whether their works were used to train a GenAI model.
b) Automated API-based workflows. Opponents claim it is infeasible to document all “covered materials” because developers “don’t know what data was used.” In reality, API-accessible fingerprinting systems have already long solved this problem: a GenAI developer could fingerprint each image at the time of ingestion for training their model via a fingerprinting service provider’s API, and later copyright owners could upload their own works to that same fingerprinting service to verify whether those images were included in the model’s training data.
AB 412 supports artists and creators by finally giving them the transparency they need to protect their copyrighted works, negotiate fair licensing terms, and hold AI developers accountable for unauthorized training. For these reasons, we respectfully ask for your “aye” vote on AB 412.
Respectfully,
Joe G. Naylor
President, Digital Media Licensing Association