Generative artificial intelligence: training data: copyrighted materials.

Assembly Member Bauer-Kahan's legislation requiring artificial intelligence developers to document and disclose their use of copyrighted training materials establishes new transparency requirements for generative AI systems operating in California.

The bill mandates that AI developers maintain comprehensive records of copyrighted materials used to train their models and provide copyright owners with mechanisms to verify if their works were utilized. Developers must implement website tools allowing rights holders to submit content fingerprints and documentation requests. When copyright owners submit verification requests with proof of ownership and content fingerprints, developers have seven days to provide a complete list of any works used. Each day beyond this deadline constitutes a separate violation.

Copyright owners whose information requests go unanswered may pursue civil action against developers, with remedies including damages of $1,000 per violation or actual damages (whichever is greater), injunctive relief, and attorney's fees. The requirements apply to all AI systems commercially deployed or made available to California users, with documentation retention mandated for the duration of the model's use plus 10 years. Developers who make their full training datasets publicly available at no cost are exempt from these provisions.

The measure interfaces with existing federal copyright protections while establishing state-level transparency obligations specific to AI development. It requires developers to track both known copyrighted materials and make reasonable efforts to identify additional copyrighted works in their training data, documenting ownership information for all protected content used.