You should definitely feel the earth moving under your feet.
By Paul Gerbino
Have you ever been in an earthquake and felt the ground move under your feet? I have not actually experienced that in real life but the way the digital landscape is constantly shifting, makes me think that the battle between AI and copyright holders has taken a major tectonic shift.
For those who may not be aware (I was not until Jacob Donnely from A Media Operator sent me this link), a class action lawsuit was filed by authors Andrea Bartz, Charles Graeber, and Kirk Wallace Johnson, who alleged that AI engine Anthropic used pirated versions of their books without permission or compensation. In the case, US District Court Judge William Alsup ruled that Anthropic’s use of copyrighted books for AI training is considered “fair use” due to its transformative nature. This is a significant decision for the AI industry’s legal defenses because there is now precedent. And there is more, the same judge found that Anthropic committed copyright infringement by storing over 7 million pirated books in its digital library, which was not justified by fair use and the potential damages that could amount to billion$.
Wait, did copyright lose? Was it a win? Was it a draw? Well I am not a lawyer but that was a major quake. In my humble but un-educated opinion, rarely does a single ruling for me illuminate the concept of “transformative” quite like Judge Alsup’s recent ruling. This isn’t just another legal case; it feels like a genuine turning point in the relationship between artificial intelligence and the human creativity it voraciously consumes. The question now looms large: Is this the moment we define the ethical guardrails for AI’s use of copyrighted material?
The Transformative Tightrope Or A Captivating Paradox
At the heart of the Judge’s ruling was an intriguing contrast. On one hand, Judge Alsup offered a significant win for AI developers. He determined that Anthropic’s use of books, and for that matter any content, to train its large language models (LLMs) was a “fair use” under U.S. copyright law. His explanation of transformative was interesting: the AI wasn’t replicating, but rather “turning a hard corner and creating something different,” making its learning process “exceedingly transformative.”
Wow! This had to be music to AI developers’ ears. They have been making this argument. They claim their AI doesn’t copy in the traditional sense; it learns, synthesizes, and generates novel content, much like a human artist draws inspiration from a vast array of existing works. Judge Alsup’s decision provides a strong legal backing to this perspective, setting precedent, potentially making access to the data/content required to fuel it more accessible. For AI, this clarity is invaluable, offering a blueprint for how future AI models might legally ingest content.
But, there is a “But.”
The same Judge delivered an equally impactful decisive blow against digital piracy. At Creative Licensing International, this has been our position all along. This was definitely a strong message to the AI world. While training was deemed fair use, there was still a shaking of the ground. The Judge stated that Anthropic’s habit of copying and storing over 7 million pirated books in a “central library” was not fair use. Judge Alsup declared that the storing and the accessible nature of that storage was not fair use. How they acquired the content and how they stored it is an issue.
Let us unpack the nuances of “fair use” here, shall we? When it came to converting purchased print library copies into their digital counterparts, that was a justified move, a different kind of fair use. The scales of justice tipped heavily in its favor, with the first factor, Purpose and Character of the Use, strongly supporting it, and the third, Amount and Substantiality of the Portion Used, lending its weight as well. The fourth factor, Effect of the Use Upon the Potential Market for or Value of the Copyrighted Work, remained squarely neutral, neither for nor against, while only the second, Nature of the Copyrighted Work, offered a slight murmur of dissent. But on balance, considering the original print copy was no more and its digital twin wasn’t being passed around like a hot potato, this was indeed fair use.
I did feel the sky tumbling down as a tremor shook the AI world when it came to those pirated copies now sitting in a centralized digital library. This was a seismic shift away from fair use.
The details here are damning: Anthropic had stolen content. They knowingly obtained content from known pirate sources like Books3, LibGen, and the Pirate Library Mirror. Judge Alsup’s pointed questioning of Anthropic’s claim that the source of their books was irrelevant underscores a crucial principle: you can’t build a business, groundbreaking technology or not, on a foundation of theft. This is a victory for copyrightholders whether authors, journalists, artists or media who have watched their works copied on illicit sites, often without a dime of compensation. It sends an unequivocal signal that while AI’s learning process might be deemed transformative, the underlying sourcing cannot be built on stolen intellectual property.
A Precedent, Not the Final Word
This judgment directly addresses fair use in generative AI and sets a significant precedent. In my non-legal opinion, it lays out a truly balanced and pragmatic view on how AI uses content for training, RAG (Retrieval-Augmented Generation), or other purposes, essentially stating that innovation should and can flourish, but not at the expense of copyright holders. The upcoming trial in December to determine damages for the pirated storage. This can reach up to $150,000 per work and will further emphasize the seriousness of this infringement. And when you multiply the 7 million works in Anthropic’s hard drive with $150,000 per work, well can you say $1,050,000,000,000! (one trillion, fifty billion!)
I believe this ruling is a big deal. It is one of the first to directly weigh in on the complex issue of fair use as it applies to generative AI and creates a definition of transformative. Before this, there was a lot of uncertainty and speculation around what transformative was and is. Now, we at least have a clearer signal from the courts.
The judgment shows a fascinating duality (I have been waiting to use duality in an article): The ruling recognizes the huge potential of AI innovation while also standing firm on the rights of copyright holders. For me it means the judge looked deeply at two very different sides of the same issue. Even though the two sides might seem to clash, both are incredibly important and need to coexist. The reasoning clearly shows that the court thought hard about both sides. It’s not about stopping AI from growing; instead, it’s about making sure that as AI gets smarter, it doesn’t do so by hurting the very people who create the content it learns from.
And Just As I Thought The Earth Stopped Moving Under My Feet… Yup, Not The Final Word
While the Anthropic ruling marks a significant moment, it’s crucial for copyright holders to understand that this is not the only case out there. After I thought I finished this article, the earth moved again. In the ongoing case of Meta versus a group of well-known writers, the judge in that case dismissed, with prejudice, a copyright infringement lawsuit.
In the Meta lawsuit, authors like Sarah Silverman, Richard Kadrey, and Christopher Golden brought a putative class action, alleging their books were used to train Meta’s LLaMA AI model without permission. Their claims included direct copyright infringement (derivative work theory which is a concept in copyright law that deals with how new creations can be legally built upon existing works), vicarious copyright infringement (a form of secondary liability where someone can be held legally responsible for another person’s direct infringement), Digital Millennium Copyright Act (DMCA) violations (1998 U.S. law that updated copyright rules to address how copyright law should function in the era of the internet and digital media, unfair competition, unjust enrichment, and negligence).
The district court dismissed most of these claims. Notably, the direct copyright infringement claim based on LLaMA being an “infringing derivative work” was deemed “nonsensical,” as the AI models themselves were not considered a recasting of the books. Vicarious copyright infringement failed due to a lack of “infringing output” from LLaMA incorporating portions of the books, and DMCA claims were dismissed because plaintiffs didn’t allege LLaMA generated and distributed copies of the books with false copyright management information. State-law claims were preempted by the Copyright Act, leaving direct infringement based on Meta’s alleged unauthorized copying of books to train LLaMA as the sole remaining claim for the plaintiffs.
U.S. District Judge Vince Chhabria ultimately ruled in favor of Meta, finding their use of copyrighted works to train LLaMA was “highly transformative” and qualified as “fair use” in that specific case which seems similar to the Anthropic ruling. It seems that Sara Silverman, et al, lacked empirical evidence showing market harm, relying instead on speculation. Meta successfully argued that LLaMA doesn’t output the actual copyrighted works and therefore doesn’t substitute for reading the original books.
And, if you thought the earth started to settle down, the after shocks came. There were important caveats and warnings from the ruling:
- Not a Blanket Approval: AI companies cannot assume a free pass from this dismissal. The decision does not state that Meta’s use of copyrighted materials to train its language models is lawful. Get Your Arguments Straight: The judge in this case said the plaintiffs “made the wrong arguments and failed to develop a record in support of the right one.” This emphasizes the need for copyright holders to get their stories straight and present compelling evidence of harm and infringement.
- Future Illegality of Unauthorized Copying: It is illegal to copy copyrighted works to train generative AI models without permission. Future rulings may rule against AI companies where copyright holders can prove AI directly copied content without authorization, especially if it can be proven to be beyond “transformative” use.
- Compensation for Creators: The courts are not naive. They know AI companies are in it for the money and expect these companies to generate “billions and billions” of dollars. With that much money AI can “figure out a way to compensate copyright holders” if using copyrighted works is necessary. What I do like is that this places the burden of developing fair compensation models right in the laps of the AI industry, not the copyrightholders.
- Limited Scope of Rulings: This ruling was limited to the specific arguments and evidence presented by 13 authors and does not establish a broad precedent that all AI training on copyrighted material is fair use.
Why These Precedents May Matter
These rulings, both the Meta and the Anthropic, may finally indicate a growing judicial understanding that while AI innovation is encouraged, it cannot come at the expense of the owners of copyright. The legal landscape, the ground we stand on, is still shifting, sometimes violently, sometimes subtly, but these early signals suggest that copyright holders have strong grounds to demand legitimate sourcing and fair compensation as AI continues to develop.
These rulings may help to establish boundaries and provide a framework for future cases and business practices. It can push AI towards more responsible development, leading to new models for licensing content and developing a more transparent AI when it comes to its data sources. It’s an important step in defining how AI can evolve sustainably, ensuring that creators are fairly compensated while technology continues to advance.
The law around fair use is still shifting, with similar lawsuits against other tech giants like OpenAI. They can either reinforce or diverge from these two findings. However, the Anthropic and Meta rulings may have drawn a line in the sand, acknowledging the transformative power of AI while simultaneously condemning the outright theft of copyrighted material.
Is this the turning point we’ve been waiting for? Is this the earth moving under our collective feet as our hearts start to tremblin? It certainly feels like it. For the first time, there is a clear judicial direction that AI’s appetite for content must be tempered with respect for copyright, especially when it comes to digital piracy of any type. Hear that content scrapers! The future of AI hinges on innovation, but also on legitimacy. And as Judge Alsup has shown the two are not mutually exclusive.
All these different rulings… Am I “just losing control, down to my very soul.” And for those not familiar with Carole King, here is a link to the song I referenced Carole King – I Feel the Earth Move (Official Audio) – YouTube. Enjoy!
About Paul Gerbino
Paul Gerbino is the President of Creative Licensing International. He is an expert in digital, content strategy, licensing, product development, advertising, and copyright protections. His expertise is noted with an exemplary track record of transforming low-performing projects into highly profitable revenue streams. Evident in creating and launching innovative digital media products and advertising programs for B2B, B2C, STM, and academic publishers. Paul is passionate about helping publishers improve their performance, productivity, and profitability in the evolving digital landscape.
