California's legislative session sees the return of **A.B. 412**, a proposed law aimed at increasing transparency in AI training data. The bill mandates that developers maintain and disclose a comprehensive list of registered copyrighted materials utilized to train their AI models. While seemingly straightforward, the **EFF** has submitted an opposition letter to the California Senate Privacy Committee, asserting that the requirements are fundamentally unworkable. ### The Impossible Compliance Burden The core challenge with **A.B. 412** lies in its demand for information that simply doesn't exist in a readily accessible format. There is no centralized, machine-readable database of copyrighted works at the **U.S. Copyright Office**. Furthermore, many copyright holders, particularly software companies, can register their works without public disclosure of the content itself. The open internet presents an even greater hurdle. Copyright information is often fragmented, unavailable, or impossible to verify. A single dataset could contain images registered with the copyright office, others licensed under **Creative Commons**, and still others in the public domain. User-generated content, like forum posts or personal photos, frequently lacks any clear indication of ownership or registration status. Essentially, the bill asks AI developers to continuously cross-reference vast quantities of online data against a copyright system not designed for such an operation. This impracticality means that if passed, **A.B. 412**'s impact would extend far beyond the major AI players typically in the headlines. ### Beyond Big Tech: Impact on Small Developers and Open Source While proponents frame **A.B. 412** as a tool for creative workers to gain leverage against large technology firms, the bill's definition of "developer" is remarkably broad. It encompasses anyone making a generative AI model available to Californians, including independent developers, open-source initiatives, non-profits, and other non-commercial efforts. Recent amendments have introduced exemptions for universities and government entities, but a significant portion of non-commercial tech work remains unprotected. Large corporations can absorb the costs of hiring dedicated compliance teams and legal counsel; smaller organizations and independent developers typically cannot. This disparity will inevitably lead to fewer opportunities for startups and new entrants, as the daunting compliance burden dissuades them from even attempting to innovate in California. ### Courts Are Already Addressing AI Copyright Questions The premise that copyright owners lack adequate recourse against AI companies is inaccurate. A growing number of federal lawsuits demonstrate that content companies are actively pursuing legal action against tech firms, both large and small, regarding AI training and fair use. These federal courts are actively grappling with complex questions surrounding fair use and transformative use in the context of AI. Some have already ruled that certain AI training activities qualify as fair use, while others continue to deliberate. Introducing new state-level regulation like **A.B. 412** while these crucial federal questions remain unresolved could create conflicting legal landscapes and undermine the benefits of a single, nationwide set of copyright rules. Ultimately, **A.B. 412** appears to be a solution in search of a problem. Existing federal law provides robust tools for rights holders to protect their interests. Rather than offering clarity or transparency, this bill introduces an almost impossible compliance burden that threatens to stifle innovation, particularly among small developers and researchers who are vital to a dynamic tech ecosystem. California has a long history of fostering both artistic creativity and technological advancement, but **A.B. 412** fails to strike a necessary balance. California residents interested in this bill are encouraged to [contact their representatives](https://findyourrep.legislature.ca.gov/).