Recap: Courts, Copyright, and the Fast-Moving GenAI Fault Line
Courts accelerate clarity, Washington shakes the guardrails, and creators brace for impact.
When I first wrote about the fair use dilemma in GenAI, I described a tension between protecting America’s creative industries and advancing technological innovation. In the weeks since, recent court rulings and swift administrative moves have started to provide more certainty, largely tilting in favor of the technology sector by expanding what fair use can cover. Yet even as the legal landscape clarifies, the fundamental tension remains: how do we support transformative innovation while still respecting and sustaining the creative economy that underpins it?
The copyright office report and the Trump administration’s response
In May, the U.S. Copyright Office released what has become its most thorough and consequential analysis yet on GenAI and fair use. Rather than issuing a sweeping declaration, the Office relied on a nuanced analysis rooted in the traditional four-factor fair use test.
The signals were clear:
Infringement is presumed when copyrighted materials are used for AI training unless a valid defense, like fair use, can be demonstrated, case-by-case.
Reproducing copyrighted works as memorized examples, even internally as model parameters, could be infringing.
Noncommercial, research-oriented uses that do not reproduce original works in outputs are more likely to be considered fair use.
Commercial uses that substitute for or compete with original works are less likely to qualify for the fair use exception.
This report advocated for a measured, case-by-case approach, balancing technological progress with the economic foundation of creative industries. It encouraged voluntary licensing markets but avoided endorsing compulsory licensing or new legislative mandates. Instead, it left room for future negotiation and adaptation, avoiding a single, blunt legal standard.
Then came the political aftershock.
Shortly after the Copyright Office shared its pre-publication report, President Trump dismissed Librarian of Congress Carla Hayden and Register of Copyrights Shira Perlmutter. This action was widely seen as a reaction to the Office’s balanced stance on broad fair use claims from the tech industry. Courts, creators, and technology companies were left without clear guidance as the report was still subject to revision or withdrawal. These moves further destabilized the policy landscape, instead of helping to resolve long-standing tensions.
The administration’s actions pushed the policy landscape toward greater industry self-regulation, favoring large tech companies due to their economic and political influence. Content creators fear revenue losses and copyright erosion. The divide between creators and tech companies widened, potentially leading to more lawsuits from publishers, artists, and rights holders. Technology firms lobby Congress to codify the right to train AI systems on copyrighted material. This unstable environment, coupled with potential future policy reversals, leaves the market uncertain. Creators, companies, and investors face shifting expectations, raising the need for Congress, the Supreme Court, or both to intervene for clear guidance.
The courts speak: recent decisions on fair use and GenAI
While the Copyright Office report tried to introduce nuance and caution into the debate, recent federal court decisions have moved quickly in the opposite direction, providing clearer (and more favorable) signals to the technology industry.
Since May 1, 2025, two major rulings from the Northern District of California have reshaped the conversation on whether using copyrighted materials to train GenAI models qualifies as fair use. These cases, Bartz v. Anthropic PBC and Kadrey v. Meta Platforms, Inc., have leaned decisively toward protecting transformative uses in AI training.
Bartz v. Anthropic PBC (June 23, 2025)
In this case, the court granted summary judgment to Anthropic, finding that using copyrighted books to train its Claude LLM was not only permissible but “spectacularly” transformative. The court compared model training to the way humans learn by reading; an analogy that effectively reframed how copyright law might apply to machine learning contexts.
Crucially, the court found no evidence of market harm to the original works, emphasizing that copyright is not meant to prevent others from learning or deriving insights. However, it drew a hard line against building digital libraries from pirated content, making clear that such uses would not enjoy the same fair use protections.
Kadrey v. Meta Platforms, Inc. (June 25, 2025)
Just two days later, Meta won a parallel case involving the use of copyrighted books sourced from “shadow libraries.” Here again, the court highlighted the highly transformative nature of training large language models and rejected the argument that simply accessing works from unauthorized sources automatically negates fair use.
Importantly, the court focused on the absence of demonstrated market harm. While it left open the possibility that stronger evidence of harm could tilt future cases against AI developers, the immediate takeaway was clear: plaintiffs will need to show concrete market impact, not just speculative threats.
Across both decisions, a few consistent messages emerged:
Transformative use is paramount. Courts are increasingly willing to see GenAI training as creating something fundamentally new, rather than substituting for original works.
Market harm remains the decisive factor. Rights holders can’t rely on general claims; they need hard evidence that AI training actually damages the market for their works.
Pirated sources are still dangerous ground. While transformative use supports fair use, relying on illicit copies or building centralized digital libraries from pirated material exposes companies to infringement risk.
Every case remains fact-specific. These victories for tech companies don’t amount to a universal green light. The legal landscape can still shift dramatically if different facts or stronger evidence emerge.
What this means: advantage tech indusry
For AI developers, these decisions provide a major, if cautious, boost. As long as the use is genuinely transformative and does not cause market harm, courts seem increasingly inclined to shield GenAI training under fair use.
For creators, these rulings underline the importance of documenting and demonstrating real-world market effects if they want to challenge AI uses effectively.
And for everyone else, from policymakers to investors, the signal is clear: the courts are moving faster than Congress or regulators, and they’re currently tilting the balance in favor of technological innovation. Yet the core tension remains unresolved: how to encourage transformative AI growth without undermining the economic viability of creative work.
Only by confronting this tension now can we protect both breakthrough innovation and the creative economy that sustains it.