On July 29, ACT | The App Association hosted a briefing on Capitol Hill, focusing on the rapidly evolving intersection of copyright law and artificial intelligence (AI). The event featured industry experts contributing the unique perspective of the stakeholders they represent:

  • Keith Kupferschmid, president & CEO of the Copyright Alliance, a trade association representing individual and corporate copyright owners
  • Ali Sternburg, vice president of information policy at the Computer & Communications Industry Association, a trade association representing communications and technology firms; and
  • Sy Damle, partner at Latham & Watkins and former general counsel at the U.S. Copyright Office.

Exploring AI and Copyright Rights

Ali Sternburg opened with a primer on U.S. copyright law and the impact of AI model training on a copyright holder’s right to reproduce their protected work. She explained that nearly everything fixed in a tangible medium, from tweets to handwritten notes, is protected by copyright, regardless of registration. Sternburg explained that this right is not absolute and, in the United States, can be qualified by a court’s finding that the use of a protected work is “fair use.” U.S. courts determine fair use on a case-by-case basis using four factors: the purpose and character of the use, the nature of the underlying copyrighted work, the amount of the underlying work used compared to the product, and whether the new use could cause harm to the potential markets of the underlying work.  She clarified that “fair use” is not merely a court defense but a fundamental right and practical safeguard for innovation and free expression, describing it as a “muscle that has to be exercised.” Sternburg highlighted the scale of copyright-implicated works in AI training and cited classic fair use scenarios (i.e., criticism, commentary, teaching, and reverse engineering), linking them to current practices like text and data mining of AI training.

Latest Insights from the U.S. Copyright Office

Keith Kupferschmid detailed the third and pre-published U.S. Copyright Office report, which focuses on generative AI training. He explained that the Office received more than 10,000 stakeholder comments, underscoring the public’s concern about the intersection of AI and copyright. The Office concluded that existing copyright law, especially the doctrine of fair use, is flexible and sufficient to address current disputes, so no legislative changes are recommended at this time.

A particularly important addition from the report is the discussion of “market dilution.” Under the fourth fair use factor —whether the new use could cause harm to the potential markets of the underlying work— the Office speculates that mass AI-generated works, while not direct copies, could flood the market and dilute the value of human-created works. However, as Kupferschmid explained, the Office acknowledges that this area is “uncharted territory,” outcomes will depend on specific facts, and the courts have yet to issue a definitive standard on when, if ever, market dilution rises to the level of infringement.

AI Training as a Tool for Innovation

Sy Damle broke down the technical framework of large language models (LLMs) in accessible terms. LLMs, whether for text, images, or music, rely on enormous datasets, some of which may be copyright-protected. Damle described how LLMs are trained: by blanking out words in sentences and repeatedly “guessing” the next word, the model’s neural networks gradually adjust weights and parameters to increase accuracy. To ensure generalization and avoid memorization, a portion of data is set aside for testing, so a well-trained model can make informed predictions on entirely new input. He stressed that while massive amounts of data are used, the intention is not to copy works verbatim but to internalize linguistic patterns and context.

Damle also clarified the process of “post-training”—using company-created, non-copyrighted conversational data to fine-tune models, making them helpful assistants. This phase, he noted, does not raise copyright issues.

Judicial Disagreements and Historical Parallels

Panelists delved into recent court cases, such as Thomson Reuters v. Ross Intelligence and two Northern District of California decisions involving generative AI: Bartz v. Anthropic and Kadrey v. Meta. Although the judges in Bartz and in Kadrey reached a decision favoring fair use in the case of AI training, they disagreed sharply on points like:

  • Whether AI learning meaningfully parallels human learning.
  • Whether a market for licensing “AI training data” is viable or legally relevant.
  • The applicability and evidentiary basis for “market dilution” arguments, (i.e., the idea that generative AI could undermine the economic value of creative works simply by overwhelming the market.)

Damle referenced technology’s “chicken little” moments, such as historical panic over photography’s impact on painters and mechanical recording’s supposed threat to musicians. In both cases, new mediums ultimately supplemented rather than displaced traditional creativity. Today’s AI, he argued, is likely to follow a similar trajectory, becoming a tool that augments creative processes rather than replacing creators entirely.

 Legislative Outlook and Panel Consensus

While panelists disagreed on key points, they all concluded that there is no immediate need for new legislation regulating AI and copyright. The panelists agreed current law, particularly the fair use doctrine, is robust and nuanced enough to adapt as the courts test and refine the boundaries. The newly proposed AI Accountability and Personal Data Protection Act, which would require explicit consent for every copyrighted work in AI training, was critiqued as operationally infeasible due to the vast scope of copyright protection and the logistical challenges of contacting rightsholders for every creative work, including everyday notes and digital content.

While some panelists acknowledged emerging licensing models and voluntary agreements, others emphasized the practical and economic limits of such strategies in the context of global AI development. With respect to U.S. competitiveness, some panelists warned that thoughtful American leadership in AI innovation should not include weakening intellectual property protections to win the global AI race.

Conclusion

The App Association’s briefing captured the developing and dynamic field that intersects copyright law, technological innovation, and creative industries. Through expert legal, technical, and policy perspectives, it became evident that sensible policymaking will require continued education, dialogue, and a careful balance between protecting creators and enabling progress. For now, the courts—and the flexibility of fair use—remain the main arbiters for reconciling the promise and perils of generative AI with the core principles of copyright. Ongoing engagement among lawmakers, creators, and technologists will be indispensable in crafting a viable path forward.