This August, the tech journalism outlet 404 Media placed an Apple AirTag in a rare book that was part of an unusual bulk order. The book’s final destination was a secret warehouse known as VGT3, “where Amazon scans and destroys books” to train its AI models.
Here, and at other AI companies’ warehouses, workers feed books into a “hydraulic powered cutting machine,” which slices off their spines. After being trimmed to fit a scanner, the books’ detached pages are scanned at high speed before being discarded or recycled.
This practice, known as destructive scanning, is standard in the AI industry, which uses books as training data for large language models such as Anthropic’s Claude and OpenAI’s ChatGPT. The scale of the practice is truly massive. Planning documents from Anthropic’s Project Panama— an “effort to destructively scan all the books in the world,”— indicated that the company wanted to convert anywhere from “500,000 to two million books over a six-month period.”
And those books were just for training Anthropic’s Claude. Because the practice is so widespread, one bookseller “estimates that the number of books being destroyed is far higher than what’s being reported — somewhere in the tens of millions.”
AI companies turn to secondhand sellers to find all these books, particularly those published before 2022, when ChatGPT and other LLMs became widely available and filled the data market with AI-generated text. Training models on AI output will not improve chatbots, so using pre-AI books guarantees a data corpus free of potentially corrupting material. Secondhand sellers had long suspected that strange, anonymous bulk orders were coming from AI companies, but were unable to prove it until VGT3 was discovered in August.
In the frenzy for new data, there is a strong possibility that rare editions with very few copies left have been destroyed. While Anthropic has insisted that they do not destroy rare or antique books, companies like Amazon have made no such promise, and using such books would give these companies a coveted market advantage.
Destructive shredding’s roots in literary piracy
Book lovers are, understandably, appalled by the destructive shredding process. The idea of literature destroyed en masse is an uncomfortable one, particularly for students of history. Numerous authoritarian regimes have destroyed books as an effort to control subjected populations. And, as Matan Grinberg of FactoryAI noted, “History seldom looks kindly on those who destroy books.”
While AI companies shred books with the much more mundane goal of gaining a competitive edge, the optics of mass destruction are not good ones. And on closer investigation, the optics problem reveals a deeper ethical issue.
The public first became aware of destructive shredding last year, when court filings from an author’s lawsuit against Anthropic revealed that the AI giant had downloaded massive databases of pirated books before legal concerns motivated a pivot to destructive shredding. Anthropic co-founder Ben Mann was particularly enthusiastic about using pirated material. Filings showed that Mann “personally downloaded” thousands of books from pirate website LibGen and praised another site called the Pirate Library Mirror, which claimed to “deliberately violate . . . copyright law in most countries.”
Last April, Meta was also sued for using pirated books to train its AI. Court documents showed Meta researchers justifying using LibGen, saying, “[T]his is what OpenAI does with GPT3, what Google does with PALM, and what Deepmind does with Chinchilla so we will do it to[o].”
When Anthropic eventually realized the legal risks of using pirated material, they turned to destructive shredding, which exploits a loophole in copyright law known as the “first sale” doctrine. This principle allows a book’s buyer to do whatever they want with it without needing permission from the copyright holder. It “allows the secondhand media market to exist” by permitting resale, but its application to destructive shredding stretches the doctrine, allowing companies like Anthropic “to avoid paying authors for their work”—which is their ultimate goal.
Walking in integrity
The backlash against Anthropic, Amazon, and Meta teaches us a valuable lesson about dishonesty. The goal of destructive shredding, and the piracy that preceded it, was to obtain books at the lowest possible cost. Doing things the right way—obtaining “direct permission from publishers and authors to use their work”—didn’t seem “practical,” according to The Washington Post’s Will Oremus. Instead, companies took illegal and barely legal shortcuts to avoid paying creators.
Both Anthropic and Meta attempted to keep those shortcuts a secret, knowing that the authors they were robbing, as well as the public, would be appalled. However, instead of asking themselves why the public would be outraged and taking that opportunity to reflect on the ethical implications of their decisions, the companies proceeded with their deceitful actions.
Those actions had a cost.
Meta, Anthropic, and Amazon faced PR disasters when, inevitably, their “sin found them out” (Numbers 32:23). Anthropic had to settle a lawsuit for $1.5 billion. The lawsuit against Meta is still ongoing, but it will likely also result in an expensive penalty. And while destructive shredding is legal for now, that could change.
The settlement ruling is not binding on other courts, which could determine that it constitutes a copyright violation. If that happens, AI companies’ prior use of piracy will “undermine [their] position,” just as one Meta researcher feared. The entire saga is a potent illustration of Proverbs 10:9—“Whoever walks in integrity walks securely, but he who makes his ways crooked will be found out.”
What should believers learn from this?
For one thing, if we’re doing something that we feel we need to hide, we should ask ourselves why and take the opportunity to repent if necessary. King David is a sobering example of what happens when we don’t. Rather than repenting after his affair with Bathsheba, David had her husband killed to cover it up (2 Samuel 11), and his actions cost him the life of his infant son (2 Samuel 12).
The Lord will not abide the presence of one who “practices deceit” (Psalm 101:7). Rather, we are called to “walk in the light” (1 John 1:7) of his commandments, living openly and righteously before God and one another.
Creating in the image of God
We should also be mindful of how AI technologies were created, how we’re using them, and what we’re using them for. God cares about means as well as ends. He gave Gideon very specific instructions for his campaign against the Midianites (Judges 7), and Saul’s improper sacrifice cost him his kingdom (1 Samuel 13:8-14). Like Gideon’s, our instructions are specific as well: we are commanded to “be imitators of God,” walking in integrity at all times (Ephesians 5:1).
AI companies’ willingness to take legal shortcuts to train LLMs demonstrates that integrity is not one of their central concerns. Unfortunately, this has been a common theme over the last few years.
In 2023, TIME Magazine discovered that an “ethical AI” firm contracted by ChatGPT was paying Kenyans less than $2 an hour to watch and label the vilest material on the internet, without warning them about the nature of the work beforehand. (Descriptions of that material are in the linked articles, so please click with caution if you want to read further about that story) In the fight over data centers, AI firms frequently use underhanded measures to hide their projects until it is almost too late for the public to stop them.
In my opinion, the ethical issues surrounding AI development should give believers pause. A tool cannot be more virtuous than its creator, and efficiency is not a kingdom value. God does not prioritize results over fairness and human dignity, and neither should we.
How and why we use AI also matters.
AI can make us more efficient, but we risk losing our God-given reasoning capabilities when we rely on it too heavily. Students who use generative AI as a shortcut for classwork are “rapidly losing the ability to read,” and studies have linked AI use with deteriorating critical-thinking skills and memory loss. As professor Tyler Jagt observed, “[O]ffloading the struggle” of learning to an LLM “deprives” us of the ability “to do any substantial cognitive work at all.”
The same principle applies to our relationship with God. God gave us a beautiful mind capable of complex thought so that we could “reason together” (Isaiah 1:18) and, in so doing, come to know him better.
He also gave us the ability to create as a way to reflect his image. J.R.R. Tolkien mused, “We make in our measure and in our derivative mode, because we are made: and not only made, but made in the image and likeness of a Maker.”
No matter how good a chatbot becomes at writing or image generation, it will never be able to reflect the image of God—or come to know him as the Creator—the way we can through our creative endeavors.
As believers, we ought to embrace the cognitive struggle of knowing God and creating in his image rather than outsourcing that process to a robot. When we do, we become more fully human as we grow closer to the Lord in whose image we were made.
How will you use your mind and your creativity to connect with your Maker today?
