Crypto news

The artificial intelligence industry has hit a paradox: the more models are trained on data created by other neural networks, the faster they degrade. This phenomenon, known as "model collapse," is forcing labs to seek increasingly sophisticated ways to obtain "clean" human material. And while it used to be about scraping websites, it has now escalated to buying up printed books and... renting human faces.

The Curse of Recursion

The term "model collapse" was formalized by an international group of researchers led by Ilya Shumailov. In their resonant paper "The Curse of Recursion," they mathematically proved that when a neural network is trained on data generated by another neural network, it loses the ability to correctly reproduce rare events and subtle patterns. Each cycle makes the model more averaged and predictable, while errors accumulate.

The scale of the problem is confirmed by the numbers. According to my analysis of web analytics data, more than 74% of new content on the internet already shows signs of auto-generation. And according to Epoch AI estimates, all usable training text on the web could run out by 2026-2032. This is why books published before 2022 have become a strategic resource—they are one of the few sources guaranteed to be free of "synthetic garbage."

Fahrenheit 451 for Anthropic

Against this backdrop, it is unsurprising that ISBNdb, a service aggregating metadata for 113 million publications, offered AI companies a specific service: purchasing books in batches of up to a million copies, followed by digitization and disposal. When the media reported on this scheme, the company rushed to remove the page with the description and stated that no deals had been made, claiming it was merely a "test of market demand."

However, independent booksellers confirm that since April 2026, anomalous purchases have been observed. Sales for one dealer grew from twenty copies to several hundred per week. Buyers are not interested in genre or value—the key is the presence of an ISBN code. Unique editions are lost irretrievably in the process: the only remaining copy may end its journey under a scanner.

Notably, courts have already begun to legitimize this practice. In June 2025, Judge William Alsup ruled that digitizing legally purchased printed books for training language models falls under the fair use doctrine. Now it is enough to keep the receipt—and the origin of the data can be documented.

Need More

In parallel, a new segment is emerging on the market—renting human appearances. Platforms like ActID and New Claw pay people from $15 to $700 for the right to use their faces as the basis for characters in AI series and advertising. In the first three months of 2026, more than 95% of the 128,000 AI mini-dramas in China were produced using artificial intelligence, creating huge demand for "living" recognizable faces.

However, this control has proven fragile. People who sold the license to their face later discovered that their digital clones were used in dishonest advertising or even political propaganda. Two high-profile cases involving British citizens who sold avatar rights to the startup Synthesia showed that their images were used for fake news in Venezuela and propaganda in Burkina Faso.

Don't Forget the Receipt

Books and faces are not all that AI companies need. Robots require examples of natural human movements. Tesla pays up to $48 per hour to people who, in motion capture suits, repeat routine actions to train Optimus machines. In 2025, more than $6 billion was invested in humanoid development.

The scheme is the same everywhere: previously, data was taken without permission; now, they pay for it. But the structure of the deal does not change—the asset irretrievably leaves the owner's control, just now with a receipt.

My conclusion: the "human data" market is becoming as commodity-based as oil or gas. But unlike hydrocarbons, losing control over one's face, voice, or text has irreversible consequences for the individual. The legalization of book buying creates a dangerous precedent: if a physical medium can be destroyed for the sake of data, what prevents extending this logic to other aspects of human identity?