In a significant turn of events last week, Anthropic, the brain behind the Claude chatbot, reached a groundbreaking settlement in a class-action lawsuit for $1.5 billion. Although this sum is substantial in the realm of copyright disputes, it represents merely a drop in the bucket compared to Anthropic’s staggering valuation of $183 billion.
The lawsuit, spearheaded by authors and publishers like Andrea Bartz and Charles Graeber, accused Anthropic of illicitly downloading millions of pirated books from shadow libraries such as Library Genesis to train Claude, thereby infringing on copyright laws. The settlement aims to compensate around 500,000 authors and publishers with approximately $3,000 for each affected work. While Anthropic did not concede to any wrongdoing, it agreed to eliminate the illegally obtained files and compensate authors, thereby averting a trial. The Authors Guild welcomed this resolution as a pivotal moment for establishing content licensing in AI development.
This case provokes critical questions regarding property rights in the era of Large Language Models (LLMs). Courts have previously determined that recombining existing texts into new forms can qualify as fair use; however, the focus of the Anthropic lawsuit was the piracy itself, rather than the training methodology. What should the legal framework stipulate about compensating authors whose works indirectly contribute to AI advancements? The resolution to this query could significantly influence not only the principles of fairness but also the future caliber of AI-generated content.
The term “AI slop” is gaining traction, referring to subpar machine-generated text that is produced with minimal human input. If human authorship becomes an untenable career path due to insufficient compensation, will LLMs lose access to a reservoir of fresh, high-quality training data? This could create a feedback loop in which AI models, trained on inferior outputs, stagnate. This scenario echoes the long-standing “access versus incentives” debate in intellectual property law: Access to a rich repository of human-written text today empowers entrepreneurs to develop robust, affordable LLMs. Yet, without incentives for human creators to continue producing quality work, the fountain of valuable training data may eventually dry up.
This situation also complicates the traditional distinction between copyright and patent law. Copyrighted works, once viewed as static, now serve as a foundation for “follow-on” innovation that builds upon the original material. In essence, copyright protections in this case have implications for AI-generated content influenced by copyrighted material, akin to how patent laws have historically governed new technologies derived from patented inventions. Thus, the “access versus incentives” paradigm applies to copyright as profoundly as it once did to patents. The Anthropic settlement underscores the urgent need for intellectual property law to adapt to the swift evolution of AI. While authors deserve compensation, stalling AI development to untangle legal disputes could stifle innovation.
With a hefty price tag of $1.5 billion, this settlement sends a resounding warning: sidestepping legal protocols could prove financially disastrous. This could deter smaller AI firms from entering the landscape, especially as similar lawsuits loom on the horizon for other companies. The precedent set here may compel developers to pursue licensing agreements or rely on public domain data, consequently inflating costs and potentially consolidating the AI sector among financially robust players like Anthropic, supported by billions in funding. Smaller startups, unable to shoulder the financial burdens of licensing or litigation, might find themselves at a disadvantage. This scenario could create a landscape where regulatory obstacles favor established incumbents. Might Anthropic’s readiness to pay such an enormous sum be a strategic maneuver to fortify its position against emerging competitors?
In a 2024 post, I speculated that AI companies, overflowing with cash, could strategically employ writers to replenish the pool of high-quality text. I posited:
“AI companies have money. Could we be headed toward a world where OpenAI has some paid writers on staff? Replenishing the commons is relatively cheap if done strategically, in relation to the money being raised for AI companies.”
The Anthropic settlement somewhat validates this notion. In an AI arms race where figures like Mark Zuckerberg invest millions to poach talent from OpenAI, $1.5 billion appears a modest price to pay for the prospect of achieving AI supremacy.
At this juncture, the Anthropic case signifies a critical inflection point. It highlights the pressing need for a balanced approach and sets the stage for how AI and intellectual property law will coexist amid an era of unprecedented technological transformation.
However, one might ponder whether, at a certain threshold, LLMs could evolve to such an extent that they no longer require human contributions. That is a future horizon that remains shrouded in mystery.
Joy Buchanan is an Associate Professor of economics at Samford University. She blogs at Economist Writing Every Day.
