California’s AB 412 Still Demands Developers Do The Impossible
Original reporting by Electronic Frontier Foundation
California lawmakers are once again considering A.B. 412, a bill designed to compel AI developers to identify and disclose all copyrighted works utilized in training their generative AI systems. On its face, the proposal appears to offer a simple path to greater transparency and fairness for creators. However, critics, including the EFF, contend that the legislation is fundamentally unworkable. The primary challenge stems from the practical impossibility of compliance: there is no centralized, machine-readable registry of copyrighted works, and the sheer volume and varied status of online content — from registered works to public domain materials and Creative Commons licenses — make comprehensive, continuous cross-referencing an insurmountable task for developers. The bill effectively demands information that often does not exist, or cannot realistically be obtained.
The unintended burden Should A.B. 412 pass, its impact would ripple far beyond the well-resourced corridors of Big Tech. The bill’s broad definition of "developer" encompasses independent innovators, open-source projects, and non-commercial entities, all of whom would face a massive, prohibitive compliance burden. While large companies could mobilize legal and compliance teams, smaller players would likely be forced to scale back or abandon their efforts, stifling innovation and consolidating power among established giants. Furthermore, the bill introduces state-level regulation into a complex legal landscape where federal courts are already actively adjudicating questions of fair use and copyright in AI. Critics argue this premature state intervention risks imposing impossible requirements without offering clear benefit, ultimately creating a costly impediment rather than a solution.
California’s A.B. 412, despite its stated goal of protecting creators, represents a fundamentally unworkable legislative approach to AI regulation. By demanding an impossible level of copyright disclosure from developers—a task that is neither technically feasible nor currently supported by existing copyright infrastructure—the bill would not only fail to achieve its aims but actively harm the very ecosystem it purports to support. It ignores the intricate realities of internet data, where copyright information is often elusive, and the robust mechanisms within federal copyright law, which are currently being tested and refined in numerous court cases. Imposing such a prohibitive and ambiguous compliance burden would inevitably stifle innovation, disproportionately impacting smaller developers, startups, and critical open-source initiatives that lack the vast legal and technical resources of larger corporations. The bill, therefore, risks solving a problem that is already being addressed, while creating new, more significant ones.
The Regulatory Horizon
The potential passage of A.B. 412 extends far beyond California’s borders, setting a precarious precedent for state-level intervention in areas traditionally governed by federal law. If adopted, it could inspire a fragmented regulatory landscape across the U.S., creating a confusing and costly patchwork of rules that would impede national AI development and global competitiveness. More critically, it risks solidifying the dominance of a few well-resourced AI giants, making it even harder for new entrants to compete and bring diverse perspectives to the field. Rather than foster a healthy balance between innovation and creator rights, A.B. 412 threatens to centralize power, slow progress, and ultimately undermine the collaborative spirit essential for responsible AI advancement. The nuanced challenges of AI and copyright demand careful consideration and a unified federal framework, not piecemeal state mandates that are ill-equipped for the task.