The artificial intelligence industry has hit a fundamental problem: high-quality data for training models is rapidly running out. The market has already responded to this shortage in an unexpected way — from wholesale purchases of printed books with subsequent disposal to licensing human faces for AI-produced series.

Model collapse and "clean" data

The key threat to future generations of neural networks is the so-called "model collapse." This term, formalized by a group of researchers led by Ilya Shumailov in May 2023, describes the degradation of AI systems trained on data generated by other neural networks. With each cycle, such models become increasingly averaged, lose touch with reality, and accumulate errors.

The scale of the problem is confirmed by the numbers. An analysis of 900,000 new pages in Google search, conducted in April 2025, showed that more than 74% of the content shows signs of auto-generation. According to Epoch AI estimates, all usable text on the internet could be exhausted as early as 2026-2032.

This is why books published before 2022 are gaining particular value — they are one of the few sources guaranteed to be free of "synthetic junk." However, access to this resource comes with legal and ethical complications. Some authors even resort to "poisoning" data by embedding hidden commands in texts. Research shows that just 250 "poisoned" documents are enough to introduce malicious patterns into LLMs.

The book business under the scanner

The practice of mass-buying books for AI training has already reached an industrial level. The case of ISBNdb, a service aggregating metadata for 113 million publications, is telling. The company offered AI labs wholesale brokerage services: buying up batches of books ranging from 1,000 to 1 million copies, with a guarantee of "purity" from AI slop and legal protection through non-disclosure agreements. Secondhand booksellers confirmed an abnormal surge in purchases starting in April 2026 — one seller's sales grew from 20 to several hundred copies per week, with buyers showing no interest in the value or genre of the publications.

After the investigation was published, ISBNdb hastily removed its promotional page and stated that it was merely a "test of market demand." However, the very logic of such a business is telling: a book on the secondary market costs $2-5, while a model trained on millions of such copies is valued in the billions of dollars. Notably, courts have already sided with AI companies — in June 2025, Judge William Alsup ruled that digitizing legally acquired books for training language models falls under the doctrine of fair use.

Renting faces and voices

A similar situation is developing in the biometric data market. In China, where more than 95% of the 128,000 AI mini-dramas released in the first quarter of 2026 were created using artificial intelligence, platforms like ActID and New Claw pay people from $15 to $700 for licensing their images. Producers select faces by categories such as "girl next door" or "brutal," while users earn passive income.

However, this market carries serious risks. British citizens who sold the rights to their digital avatars to the startup Synthesia found their "clones" in propaganda videos supporting the dictatorial regimes of Venezuela and Burkina Faso. As lawyers rightly note, licensing terms are often so vague that it is impossible to control the final use of one's appearance.

My verdict

We are witnessing the formation of a new type of commodity market, where human experience — texts, faces, voices, movements — becomes as much a resource as oil or rare-earth metals. The pattern is the same everywhere: first, data was taken without permission, now people are paid for it, but the asset still irrevocably slips out of the owner's control. The only difference is that now you have a receipt. The question is how long this illusion of control will last before regulators realize the need for fundamentally new rules of the game.