The current atmosphere in the AI development community is defined by a shift in legal warfare. For the past two years, the primary battleground has been the question of whether training a model on copyrighted data constitutes a violation. However, a new and more dangerous argument is emerging in the courts. It is no longer just about whether a model stole a specific sentence or a particular melody, but whether the very existence of a generative tool destroys the economic viability of a creative profession. This shift moves the conversation from the act of copying to the act of competing, creating a tension that could redefine the boundaries of intellectual property for the next century.
The Legal Architecture of Market Dilution
The Electronic Frontier Foundation (EFF) has stepped into this fray by filing amicus briefs in two pivotal cases: Concord Music Group, Inc. v. Anthropic PBC and In re Mosaic LLM Litigation. At the heart of these filings is a direct challenge to the market dilution theory. This theory, championed by copyright holders, posits that the creation of generative AI tools cannot be considered fair use because the resulting ecosystem floods the market with AI-generated content. The argument suggests that because these tools enable the proliferation of works that compete with human creators, the tools themselves are the instruments of market destruction.
Under this logic, copyright holders are asking the courts to expand their reach. They argue that they should have the power to control not only direct copies of their work but also non-infringing works created by others using AI. Essentially, they are requesting a modification of 300-year-old copyright principles to prevent the market erosion caused by AI. The goal is to ensure that the act of building a model is seen as an inherently infringing act if the resulting tool allows users to generate content that mimics the style or utility of the original copyright holder's work.
The Distinction Between Infringement and Competition
This is where the EFF draws a sharp line. The foundation argues that the market dilution theory fundamentally contradicts the constitutional purpose of copyright law. Copyright was designed to promote the progress of science and useful arts by granting limited monopolies, not to protect a specific industry from the natural evolution of technology. The EFF asserts that the law is intended to punish copyright infringement, which is the unauthorized reproduction of a specific expression, rather than market competition, which is the act of offering a competing product or service.
History provides a repetitive pattern of this exact panic. In 1906, critics claimed that automatic pianos and phonographs would destroy the art of musical composition. Later, the advent of the camera sparked fears that portrait painters would be rendered obsolete. In both instances, the technology did not kill the art; it transformed it. The camera did not destroy painting but instead pushed it toward abstraction and impressionism while creating entirely new fields like photojournalism. The EFF argues that LLMs are the modern equivalent of these general-purpose tools.
Another critical precedent is the 1980s battle over Video Tape Recorders (VTRs). Copyright holders argued that VTRs would destroy the movie industry by allowing people to record shows. However, the U.S. Supreme Court recognized time-shifting as a non-infringing use, warning that the law should not be rewritten simply to react to a new technology. The EFF positions LLMs in this same category of general-purpose tools. AI does not just generate lyrics or images; it assists in coding, research, and complex problem-solving, amplifying human creativity rather than simply replacing it.
From a technical perspective, the EFF points out that as the volume of training data increases, the influence of any single piece of data on a specific output diminishes. This suggests that the likelihood of a model producing a work that directly infringes on a specific copyright is low, further separating the act of training from the act of infringement.
If the courts accept the market dilution theory, the result would be a functional veto power for copyright holders. They would not need to prove that a model copied their work; they would only need to prove that the model's output competes with their style or genre. This would effectively grant ownership over tropes, styles, and general modes of expression, creating a legal environment where AI developers could be sued for the mere possibility of competition. Conversely, if the court follows the EFF's reasoning and maintains the fair use principle, the legal focus remains on whether a specific output is a direct reproduction of a protected work.
This distinction determines whether the future of AI is built on a foundation of open competition or a series of restrictive licenses controlled by a few legacy rights holders.
