Books are burning where no one’s watching

Book burnings in the olden days were displays of power, eerily ceremonial in nature. People would watch on as the flames climbed up to the sky, fueled by literature deemed threatening or disparaging towards the dominating authority. 

But unlike those grandiose displays of the past, today’s book “burnings” have taken a new form; this time, the destruction is more covert, more surgical. Artificial intelligence companies behind generative artificial intelligence (GenAI) are actively destroying books to further their dystopian agenda of keeping the populace dumb and complacent.

A few weeks ago, this modus operandi of “destructive scanning” of books surfaced in two separate events. First, the International Standard Book Number database (ISBNdb) had offered its catalogue of books to AI companies to train GenAI models. In a now-deleted news page, ISBNdb had boasted its collection, saying “the world’s best AI training data is sitting on a shelf”. On July 30, they walked back on their claims, citing the service was a “test of market interest”.

Second, last month, unsealed court documents from the biggest settled copyright lawsuit against Anthropic—the AI company behind Claude, a GenAI model—trended on the Internet at roughly the same time. The documents revealed that the company has engaged in “destructively scanning all the books in the world” for training data since 2024. Codenamed Project Panama, Anthropic has been buying out books in bulk, converting them into private digital copies to serve as training data, and destroying the physical copy afterwards. They contracted Datamation, a digital transformation and document scanning vendor whose name is conveniently left out of mainstream media news reports, to destructively scan two million books within six months. Following the unsealing, these documents, especially footage of a hydraulic blade slicing off the spines of these books, garnered the outrage of the Internet.

These events may not be directly correlated, but they are not isolated cases. In fact, they are all part of GenAI’s corrosive learning process at a destructive scale. Models like Claude, ChatGPT, and Gemini derive their power from enormous datasets obtained through numerous means such as scraping and crawling the Internet, generating synthetic data using statistical models, and buying aggregate data from data companies. By iterating through these data in “training”, the models learn to associate and recognize patterns within the given dataset, minimize mistakes in predictions compared to actual values, and optimize their parameters. Thus, these models generate outputs on a query based only on what it is trained upon.

During the earlier years of the AI boom, clean human data was abundant for AI models to utilize. However, since last year, more than 50% of data on the Internet is polluted by GenAI content, and because these models remain largely dependent on the web for training data (much to the dismay and disgust of its users), the pollution has been infecting GenAI models trained on it. In turn, it has subtly led to subpar performance, loss of variance, and generic results. In other words, GenAI models have been eating their own outputs and producing worse iterations, which are again fed into more models, fueling the phenomenon known as model collapse.

This plateau in GenAI development made AI companies desperate for fresh data. So, they turned their attention to books published before 2022, free from pollution by GenAI. Rare literature, old classics, limited copies—Anthropic did not discriminate when it came to pulverizing them in bulk, and ISBNdb did not hold back during the time they offered their catalogue as a service. In fact, many AI companies and book databases have hopped on the bandwagon of using physical literature as training material; only Anthropic and ISBNdb made the news when they were caught.

What were once volumes of information and products of human imagination since time immemorial are being desecrated by these AI companies merely for data. Destructively scanning for books is an insult to the human minds that synthesized masterpieces from pure experiences, and the end goal of harvesting GenAI-free text makes it all the more affronting towards the collective knowledge humanity spent building for thousands of years. These tomes of knowledge are lost forever for something as surface-level as GenAI.

The damning part of all this does not even lie in the fact that it is still happening as of writing this article. In the copyright lawsuit against Anthropic, the federal judge ruled that destructively scanning books is considered fair use under copyright, simply because the company paid for them. To put it plainly, buying physical copies from public booksellers, cutting off their spines, scanning the pages into digital copies for private use, and pulping the remains is considered “transformative” and therefore legal.

In other words, the law fully supports the legal framework Project Panama takes advantage of to deprive the public of essential books in favor of training GenAI models, as long as the books were legally obtained. The ruling from this lawsuit became the permission AI companies needed to further develop their models by any means necessary, even if it means sacrificing entire cultures for efficiency.

The commodification of books and their degradation from cultural symbols to GenAI dataset fuel reveal the endgame of the entire GenAI industry. In fact, it was succinctly summarized by Sam Altman, the CEO of OpenAI (the company behind ChatGPT), during a summit last March 2026: “We see a future where intelligence is a utility, like electricity or water, and people buy it from us on a meter.” These companies intend to monopolize the intelligence of the human race, making it only accessible through their own GenAI models.

Destructive scanning of books is a step towards that goal that benefits them in two ways. The models improve in performance due to fresh sources of data, while they also destroy copies of books and prevent the general public from accessing them. They are replacing the existence of collective intelligence with an artificial one, forcing dependence upon a single source of knowledge.

And that spells disaster for the majority of the population. Since last year, mounting scientific evidence confirms that GenAI makes people dumber through prolonged use; human brains are less active and more forgetful when they offload cognitive tasks such as reading and writing. The longer users engage with GenAI, the more they accept its models’ output without any cognitive exercise, rejecting hard-earned intuition in favor of frictionless answers.

Human minds that freely accept any idea given to them is dangerous, and it is exactly what these powerful figures want out of the public. Because GenAI models play into the biases found within their datasets, they can be influenced by governments and companies to promote propaganda through their preferred mechanical mouthpiece. Plus, without collective knowledge and active thinking skills to rely on, it would become more difficult for people to corroborate claims, challenge ideas, and consider other perspectives. Instead, they would be feeding off whatever warped information more powerful figures want those below them to hear.

This vicious cycle of cognitive numbness only fuels the prevalent literacy crisis in the country, especially among Filipino youth. Even without GenAI in the picture, many students already struggle with basic reading and proficiency across all levels—even as they enter adulthood—and the national education system is already designed in a way that consistently fails most of its constituents. Now, with the Philippines among the fastest countries in Southeast Asia to adopt GenAI in the education sector, more children will become vulnerable to the slippery slope of GenAI dependence and cognitive decline due to the unbearable duress of their education exacerbated by this poor “system”.

Aside from schools, libraries and bookstores are also at risk. After all, they are the main target of AI companies when it comes to procuring books for destructive scanning. According to booksellers, they have seen an influx of orders on books with ISBNs, as well as rare and out-of-print material, and because they are legal purchases, they have no choice but to comply, even if the end product is solely for a machine incapable of human thinking. Despite the lucrative short-term financial gain for these shops from these orders, they are depleting the value of these bastions of thought and progress at a terrifying pace. If this trend continues, bookshops worldwide would close down, libraries would end up abandoned, and reading itself would be reserved for the wealthy few.

Simply put, AI companies are quietly erasing spaces and artifacts that promote intellectualism, replacing them with systems that perpetuate stupidity. And as long as perverted methods like destructive scanning are not only allowed but supported by the law, then, much like the model collapse GenAI suffers from today, nuanced human thinking would homogenize into a dense mental fog that can be molded by any external influence.

Thus, it is vital that GenAI usage should be stopped. At an individual level, people should refuse to offload work to GenAI, even for simple tasks such as writing mundane details or articulating thoughts, and, instead, devote time and effort to sharpening the mind with essential literacy skills. At a community level, initiatives like hyperscale AI data center construction and bulk buying of literature must be met with vehement opposition. Governments and organizations should protect books and libraries from the ongoing desecration of these cultural artifacts, and thoroughly penalize AI companies for the destruction of art and literature.

Destructive scanning is this age’s form of book burning. But instead of a raging fire, it is a cold hydraulic blade. Instead of just leaving ashes behind, it is storing away digital copies of books before pulping the physical ones. And instead of a powerful figure, the one standing to benefit from these “burnings” is generative artificial intelligence with its unceasing hunger for fresh human output to convert into data.

Whether by rageful flame or unfeeling blade, what is at stake remains the same, regardless of the method: Human thought is being suppressed and censored by AI companies in real time. Humanity has already witnessed what happens when a dominant authority controls a monopoly of knowledge. So, amid all that intentional obliteration today, only one question remains.

How long will we wait before the last book is “scanned” away, never to be seen again?