California AB 2013 — Generative AI: Training Data Transparency
Requires developers of generative AI systems made available to Californians (covering systems released since January 1, 2022) to publish, before release or substantial modification, a high-level summary of training datasets: sources, number and types of data points, copyright/IP status, whether data was purchased or licensed, and whether personal information was included. Exemptions for security, aircraft-operation, and federal national-security systems.
Enacted (chaptered Sep 28, 2024); operative Jan 1, 2026. xAI constitutional challenge pending on appeal in the Ninth Circuit after denial of a preliminary injunction.as of 2026-01-01primary source ↗
“The government’s consultation, along with its preferred policy option on AI and copyright, fails to address the real issue. The UK’s gold standard copyright law is already clear; what is missing is robust enforcement and transparency requirements that empower creative businesses to assert their rights.”
News Media Association (UK); statement of chief executive Owen Meredith. Coded support relative to the transparency ask; the quoted sentence criticises the government's preferred opt-out option.
“Big Tech needs to pay for the creative and research content they hoover up to train AI, just as they pay for their electricity and other normal costs of running a legally compliant business. We urgently need transparency regulations to lift the lid on AI usage”
“Yesterday techUK wrote to the Secretary of State for Science, Innovation and Technology Peter Kyle MP to reiterate our position that the Data (Use and Access) Bill is the wrong legislative vehicle to address the complex issue of AI and copyright.”
“Despite actively acknowledging that creative work is being stolen at scale, the Government chose once again to remove transparency provisions your Lordships provided, allowing the tech sector to continue to rob the creative industries blind.”
“This bill helps to achieve that commonsense goal by creating an administrative subpoena process that allows copyright owners to obtain information from AI companies about the works used to train their models. As the Guild has long recognized, this kind of transparency is essential to giving authors the ability to enforce their rights.”
“Artists deserve to know when their works are used and determine the parameters – the Transparency & Responsibility for Artificial Intelligence Networks (TRAIN) Act empowers rightsholders with valuable insight into which specific inputs AI companies are copying to train their models. Just as importantly, it also ensures a clear path to the courts when authorization has not been given.”
“We strongly reject any claim that the Code of Practice strikes a fair and workable balance or that the Template will deliver “sufficient” transparency about the majority of copyright works or other subject matter used to train GenAI models.”
Joint statement of 40 European and international creative and cultural sector organisations incl. CISAC, IFPI, European Writers' Council, STM.