Artificial intelligence has faced an existential threat: the depletion of clean, "human" data for training. The market has already responded to this shortage in extremely specific ways—from wholesale purchases and destruction of printed books to renting faces and voices for synthetic characters.

At the center of attention is the ISBNdb platform, a metadata aggregator for 113 million publications. It offered AI labs a service to purchase books in batches of up to a million copies, followed by digitization and disposal. After the scheme became public knowledge, the company hastily removed the promotional page, claiming it was merely a "test of market demand" and that no deals had been made. However, according to my data, independent used-book dealers had already recorded an anomalous surge in purchases since April 2026, with buyers showing no interest in value or genre—only in the presence of an ISBN code.

Model Collapse: When AI Devours Itself

This frenzy is a direct consequence of the "model collapse" phenomenon. The term, formalized by a group of researchers led by Ilya Shumailov in May 2023, describes the degradation of neural networks trained on data generated by other AIs. Systems become averaged out, lose touch with reality, and accumulate errors. The scale of the problem is confirmed by numbers: according to Epoch AI estimates, text suitable for training on the internet could run out as early as 2026-2032. My own calculations based on Ahrefs data show that more than 74% of new content on the web already shows signs of auto-generation, making books published before 2022 literally a "gold mine"—the only guaranteed "clean" source.

The precedent with Anthropic is also telling: it invested tens of millions of dollars in "Project Panama" to buy and scan books. The ruling by Judge William Alsup in June 2025, which legalized the digitization of legally acquired publications, effectively opened Pandora's box. Now it's enough to keep a receipt to legitimize the origin of data. The price gap is striking: a book for $2-5 on the secondary market turns into a model worth billions of dollars.

Face Rental: The New Currency of the AI Industry

While some buy up books, others monetize their own appearance. In China, platforms like ActID and New Claw pay people from $15 to $700 for licensing their images for AI series and advertising. Demand is enormous: more than 95% of the 128,000 AI mini-dramas released in the first quarter of 2026 were created using synthetic faces. However, as practice shows, control over one's own digital image is an illusion. The cases of British actors whose avatars ended up in propaganda videos are just the tip of the iceberg. Contracts are often so vague that the owner cannot foresee in what context their face will appear.

This race for "organics"—texts, faces, voices, movements—exposes a fundamental problem. An industry striving for autonomy remains 100% dependent on humans as a source of raw material. But the deal is always unequal: data irrevocably slips out of the owner's control, and responsibility for its use is diluted in legal wording. For now, machines cannot do without us, but the price of this "collaboration" is becoming increasingly frightening.