Regulatory Intervention and Artificial Inspiration: Examining Google’s AI Governance Proposal on Copyright Protections

Similar to other major AI developers, Google has been facing several lawsuits that challenge the tech giant’s alleged use of copyrighted materials to train its AI models. Against that backdrop, Google’s white paper on AI policy, A Pragmatic Approach to AI Governance in America, appears to present a balanced approach to facilitating continued innovation on one hand, while addressing public concerns on the other. However, at least with respect to copyright protections, Google’s arguments seem to favor AI development over the interests of recording artists, songwriters, authors and other rightsholders whose intellectual property may be used to train AI systems.

The Two-Tier Approach to AI Governance

Google addresses two areas of AI as separate and distinct, requiring substantially different approaches to their governance and regulation. For “frontier AI” models, developed at the cutting-edge of the technology, Google recommends establishing an industry-funded entity that would operate under federal oversight to mitigate potential risks to national security. For models released to the general public for individual and commercial use, considered “widely-deployed AI,” Google encourages “figuring out how our existing laws apply to this new technology,” noting further that “we don’t need to reinvent the wheel.”

Upon closer review, Google’s position on widely-deployed AI could be interpreted as a sliding scale of proposed regulatory intervention. On one side are new and enhanced policies in areas affecting child safety, information integrity and privacy rights. On the other side are areas such as energy infrastructure, workforce policy and copyright protections, where Google largely refers to existing frameworks to address legal questions stemming from AI.

On Copyright Protections and Enforcement

To begin the discussion on widely-deployed AI, Google offers what appears to be a common-sense consensus: “People on the left and the right agree that if something is illegal to do without AI, it’s illegal to do with AI.” While the statement may be attempting to invoke broad appeal, its underlying logic is debatable. First, in the absence of any data, citation or references, the asserted bipartisan agreement would be anecdotal at best. Separately, the subsequent phrase (“if something is illegal to do without AI…”) could be interpreted as suggesting that actions performed by AI should be considered legal if the same would be true if they were performed by a human. Not only is this notion currently being challenged in courts, but Google itself acknowledges later in the same paragraph, and throughout the paper, that AI raises novel questions concerning the applicability of current legal concepts to it.

On training data, Google argues that “using publicly available web data for training models is a transformative, non-expressive use — like an art student taking inspiration from walking through a gallery.” The comparison in this analogy may be overlooking important differences between human inspiration and machine learning. A student who visits an art gallery acquires knowledge, influences and techniques through observation. The student does not make perfect copies of every artwork, analyze and aggregate details from each one and potentially generate an unlimited number of derivatives upon request. Whether the alleged use of copyrighted works as model inputs ultimately constitutes infringement is a question for the courts, but drawing parallels between the two activities seems hard to reconcile.

In case of copyright challenges to AI outputs, Google suggests that “the appropriate mechanism is through established notice-and-action frameworks that use standard reporting and takedown mechanisms to remove infringing content.” This statement does not seem to account for the immense scale at which generative AI is capable of producing. Ongoing lawsuits involving other AI platforms allege that such models ingested tens of millions of works as training inputs, enabling them to generate derivative outputs at an unprecedented scale. Requiring rightsholders to monitor a continuous stream of AI-generated content, identify potentially infringing works and submit individual takedown requests would likely increase the burden of enforcement significantly for creators, while essentially shifting responsibility for the output onto users of the platform.

Conclusions and Implications for Rightsholders

Throughout the paper, Google promotes its approach as “pragmatic,” “balanced,” “evidence-based” and “data-driven.” Yet in the context of copyright protections, the paper seems to rely primarily on analogies and assumptions rather than empirical analyses. An evidence-based approach should recognize the imbalance these technologies introduce into existing frameworks while also considering how recent licensing deals and partnerships may be shaping market expectations. Likewise, a pragmatic approach should explore solutions that allow for continued AI innovation while ensuring that rightsholders share in the underlying economic value created by their works.

These issues extend far beyond creative rights, encompassing both current creator compensation and the long-term value of underlying intellectual property assets. As litigation, licensing deals and industry practices continue to shape market expectations, rightsholders may benefit from independently validating and reassessing their contractual arrangements to determine whether they adequately address emerging forms of use and monetization of their works.

Withum’s Theatre and Entertainment Services Team is passionate about the industry and closely monitors market trends to stay informed of evolving business, legal and regulatory conditions affecting rightsholders. Our Royalty and Intellectual Property Audit Services help clients gain meaningful insights into their portfolio, recover underpaid royalties and ensure their creative works are properly credited for the value they create.