Article by Christina Vaughan, 16th September 2026
In a previous article, I asked how worried we should really be about artificial intelligence, following warnings from some of the very people building it that advanced AI could ultimately pose an existential threat to humanity.
Now, I find myself considering a different but connected question. If AI represents the future, why are some of the world’s most advanced technology companies so hungry for the creative work of our past?
Project Panama was Anthropic’s extraordinary attempt to acquire and digitise millions of physical books. The books were purchased, their bindings removed and their pages scanned before the physical remains were recycled. For anyone who loves books, the smell, the touch and that peaceful sense of opening one and allowing your mind to disappear into it, the image is deeply unsettling.
But once I got beyond my instinctive horror at the destruction, another thought began to trouble and fascinate me. We live in an age increasingly flooded with synthetic content, yet one of the world’s most sophisticated AI companies went to enormous expense to acquire human-written books.
That tells us something profound. Human-created content is not becoming less valuable in the age of AI. It is becoming strategically precious.
What was Project Panama?
According to documents unsealed as part of a copyright lawsuit against Anthropic, Project Panama began in early 2024 as an ambitious and deliberately discreet programme to acquire and scan books on an enormous scale.
The Washington Post reported that Anthropic spent tens of millions of dollars buying millions of books, many sourced through booksellers and other suppliers. Their spines were removed so the pages could be scanned quickly and efficiently, providing material that could support the development of Claude and Anthropic’s wider AI systems.
The details emerged from more than 4,000 pages of court documents. Together with filings in other copyright cases, they have offered an extraordinary insight into the race among some of the world’s largest technology companies to acquire the vast quantities of material needed to develop increasingly powerful generative AI.
Books were particularly valuable because they contain something very different from much of the material found online: sustained thought, carefully constructed language, specialist knowledge, complex arguments, long-form narratives and the cultural record of human experience and emotion.
One Anthropic co-founder reportedly suggested that books could teach AI models “how to write well,” rather than merely reproducing low-quality internet language.
That is telling.
Generative AI may appear to create something new, but it does not develop in a cultural vacuum. Its capabilities are built upon vast quantities of existing human expression: the books we have written, the photographs we have taken, the music we have composed and the knowledge we have painstakingly recorded.
If synthetic content were an adequate substitute for human creativity, why would Anthropic go to the extraordinary expense of purchasing and scanning millions of human-written books?
And, in my own world of visual content, why are technology companies continuing to commission so much new human-created photography?
AI still needs human culture
I have spent more than three decades working with human-created imagery, commissioning photographers, licensing intellectual property, securing appropriate releases and ensuring that models and contributors are fairly represented and properly paid.
From that perspective, Project Panama is not simply a story about books. It exposes a much larger truth about the value of creative content in the age of AI.
AI systems still need material created by people. They need our knowledge, imagination, experiences, cultures and differences. They need work shaped by original human choices, rather than material generated from the outputs of earlier machines.
This becomes increasingly important as the internet fills with synthetic material.
I am not saying that synthetic data is inherently bad. Properly designed and controlled, it can be enormously useful, particularly where real-world information is scarce, sensitive or difficult to obtain. But that is very different from an uncontrolled digital environment increasingly populated by unattributed, repetitive and machine-generated content.
If AI systems increasingly learn from the outputs of other AI systems, errors can be repeated, stereotypes reinforced and cultural specificity gradually flattened. Material may appear convincing while becoming increasingly detached from a reliable human source.
The great paradox of generative AI may therefore be that the more synthetic content it produces, the more valuable dependable human-created source material becomes.
The legal distinction matters
Before going further, an important distinction must be made.
Project Panama involved physical books that Anthropic had legally purchased. The company’s subsequent $1.5 billion settlement concerned a separate collection of digital books obtained from pirate libraries.
These two methods of acquisition formed part of the same wider pursuit of training material, but they should not be conflated.
In June 2025, US District Judge William Alsup ruled that using lawfully acquired books to train Anthropic’s AI models constituted fair use in that particular case. However, he found that acquiring and retaining millions of pirated books for a central digital library raised a separate legal issue that was not excused by their possible later use for AI training.
Anthropic subsequently agreed to a $1.5 billion settlement concerning more than 482,000 pirated works. The settlement received final Court Approval in July 2026.
This was a decision made under US fair-use law. It should not be interpreted as establishing that equivalent activity would automatically be lawful here in the UK, in the European Union or elsewhere.
But the legal question is not the only question.
Something can survive a particular legal test and still raise legitimate concerns about transparency, consent, attribution, environmental impact and the fair treatment of creators. Lawfully purchasing a physical copy of a book may answer the question of who owns that particular copy, but it does not automatically resolve every ethical question about how the intellectual and creative work contained within it should be used.
Provenance is part of the product
At Cultura Creative, these questions are not theoretical. We ask them every day.
Where did this image come from? Who created it? Who appears within it? Were the contributors properly informed, contracted and paid? What permissions were granted, and for which uses? Can every asset be traced through a dependable chain of rights?
Our content is commissioned and created specifically for legitimate commercial licensing. We work with known contributors, secure appropriate model and property releases and intentionally build diversity into both production and representation. We also create the detailed metadata that makes visual material genuinely useful to the organisations licensing it.
For us, provenance is not an administrative detail attached to a dataset. It is part of the dataset’s value.
I saw this throughout more than two decades of building Image Source into an internationally successful stock photography business, and I see it again today at Cultura. Content with a clear chain of rights is not only the more responsible choice; it is a safer and more valuable commercial asset.
It reduces legal and reputational risk. It provides more reliable metadata and greater consistency. It allows better control over representation and bias. It also gives the customer confidence about how, where and why the material can be used.
This is not about Cultura sitting on its high horse or claiming that every complex question surrounding AI has a simple answer. It is about recognising a commercial reality.
The next phase of AI will not simply require more data. It will require better data: material that is relevant, diverse and accurately described, with origins and permissions that can be properly explained.
The future cannot be built responsibly from material whose ownership, provenance and intended use nobody can adequately account for.
Human creativity as infrastructure
Project Panama can be viewed as a story about technology physically consuming the products of human culture. But it also tells us something more hopeful.
Human creativity has not become obsolete in the age of AI. It has become foundational infrastructure.
The question is not whether AI will continue to learn from human-created work, it must. The real question is whether the people and organisations building it will acquire and use that work transparently, legally and responsibly.
Project Panama demonstrates that authentic human-created content is becoming more valuable as AI advances. But provenance, permission and ethical sourcing must now become as important as scale.
The companies that understand this will not only build more trustworthy AI. They are also more likely to build better AI.
Christina Vaughan
Founder & CEO, Cultura Creative
16th September 2026