Sponsors (2)
The bill strengthens copyright owners' enforcement rights and transparency around AI training uses but does so by imposing new notice, reporting, and penalty requirements that raise compliance costs, legal risk, and potential exposure of proprietary dataset information for developers and businesses.
Copyright owners can sue and obtain statutory penalties when their registered works are used to train models without required notice, giving rights holders a clearer enforcement path and potential compensation.
A public registry/database will list notices of works used in training datasets, increasing transparency about which copyrighted works power models and making it easier for rights holders and the public to discover use.
Rights holders receive at least 30 days' advance notice before a model trained on their works is commercially used or released, giving creators time to respond or seek remedies before market entry.
Developers and companies must prepare detailed dataset summaries and file notices before commercial use, raising compliance costs and likely delaying product launches—burdening especially startups and small teams.
Large civil penalties (minimums of $5,000 per failure and up to $2.5M per year) create substantial financial and legal risk for model builders that may chill AI innovation and raise costs for consumers.
Maintaining a public database with dataset links and details risks exposing proprietary dataset URLs or sensitive business information, harming competitiveness and possibly security for dataset holders.
Based on analysis of 2 sections of legislative text.
Requires advance or near-term notices to the Register of Copyrights listing copyrighted works in datasets used to train or release generative AI models and creates a public database and civil remedies for failures to file.
Official title: Require a notice be submitted to the Register of Copyrights with respect to copyrighted works used in building generative artificial intelligence models, and for other purposes.
Introduced February 10, 2026 by Adam Schiff · Last progress February 10, 2026
Requires anyone who trains or releases a generative AI model to notify the Register of Copyrights with a detailed list of copyrighted works in the training dataset (and dataset URLs if publicly available). Notices must be filed 30 days before commercial use or release (or within 30 days after required regulations for models already released). The Copyright Office must write regulations, run a public online database of notices, and may collect penalties paid to offset its costs. Civil remedies for copyright owners are created, including minimum penalties of $5,000 per failure with an annual per-person cap of $2,500,000.