AI firms are seeking "organic" data by buying pre-2022 books and licensing faces amid fears of model collapse. ISBNdb, a service aggregating metadata on 113 million titles, pitched purchasing books in batches of up to one million for digitization and disposal, but removed the page after media reports, calling it a "market demand test." Platforms ActID and New Claw pay people from $15 to $700 for the right to use their faces in AI series and ads. "Model collapse" describes AI degradation from training on machine-generated data, formalized by Ilya Shumailov's team in May 2023. In April 2025, Ahrefs found materials with signs of auto-generation exceeded 74% among 900,000 newly indexed pages. Epoch AI estimates web text suitable for training could run out between 2026 and 2032. Books published before 2022 are valued as sources free of synthetic noise. Some authors use "data poisoning," hiding characters in text; a study found an LLM can be covertly trained on malicious patterns with as few as 250 documents. In August 2024, authors Andrea Bartz, Charles Greibner, and Kirk Wallace Johnson filed a class-action lawsuit against A.