Crypto news

The artificial intelligence industry has hit an unexpected paradox: the more models are trained on synthetic data, the more valuable "organic" sources become. This isn't just about texts, but also human faces, voices, and even movements. New, sometimes shocking schemes are taking shape in the market—from mass book buyouts to renting appearances for AI-generated series.

Model Collapse: When AI Eats Itself

The key issue is the so-called "model collapse." When a neural network is trained on data generated by another neural network, it degrades: responses become averaged out, and the connection to reality is lost. The term was first formalized by researchers led by Ilya Shumailov in May 2023. They mathematically proved that regularly mixing in synthetic data causes the model to forget rare events and accumulate errors.

The scale of the problem is backed by numbers. In April 2025, Ahrefs analysts examined 900,000 new pages in Google search—more than 74% showed signs of auto-generation. According to Epoch AI estimates, text suitable for training on the internet could run out between 2026 and 2032. This is why books published before 2022 are becoming a strategic resource—they are guaranteed to be "clean" of machine-generated noise.

The ISBNdb Scheme: Millions of Books Under the Scanner

ISBNdb, a service holding metadata for 113 million publications, offered AI companies a service to purchase books in batches of up to a million copies. The scheme is simple: books are bought on the secondary market, scanned, and destroyed. After the investigation was published, the company hastily removed the promo page, claiming it was just a "test of market demand." However, independent booksellers confirmed an abnormal surge in purchases starting in April 2026: one dealer's sales jumped from 20 to several hundred copies per week.

Notably, buyers weren't interested in the books' value—only the presence of an ISBN. Unique editions are irretrievably lost under the scanner. Meanwhile, the deal's economics are absurd: a book worth $2–5 turns into a model valued at billions of dollars. A precedent has already been set: in June 2025, Judge William Alsup ruled that digitizing legally purchased books for AI training falls under the fair use doctrine. Just keep the receipt.

Face Rental: China's New Market

A similar situation is unfolding with human appearances. In China, where over 95% of the 128,000 AI mini-dramas released in the first quarter of 2026 were created with AI, platforms like ActID and New Claw pay people from $15 to $700 for licensing their faces. Users upload photos in special studios, and producers pick archetypes—from the "girl next door" to a "brutal" character.

This model didn't emerge out of prosperity: traditional drama production is shrinking, advertising budgets are falling, and luxury brands are shifting to AI content. However, risks remain. Beijing lawyer Ile Deng notes that licensing terms are often so vague that it's impossible to tell who ends up using the appearance and how. There have already been cases where sold avatars were used in political propaganda—for example, in fake news linked to the Maduro regime in Venezuela.

What's Next?

The scheme is the same everywhere: data used to be taken without asking, now it's paid for. But the essence hasn't changed—the asset irreversibly slips out of the owner's control, just now with a receipt. Books, faces, voices, movements—all of it is turning into raw material for training models. The market is only taking shape, and so far it works in favor of corporations, not data providers.

My takeaway: We are witnessing a fundamental shift in the data economy. The AI industry, seeking to avoid "model collapse," is creating a new class of assets—"clean" human data. But without clear regulation, this market risks turning into digital servitude, where individual rights are diluted by corporate interests. Investors should closely watch legal precedents in this area—they will determine the future of the entire industry.