Data scraping for artificial intelligence training refers to the systematic collection and digitization of physical or digital content to feed large language models (LLMs). Recent investigations, including reports from 404 Media and Ars Technica, have highlighted the physical destruction of rare and copyrighted books within corporate facilities to facilitate high-speed scanning and ingestion into AI training pipelines. This practice raises significant questions regarding intellectual property rights, the preservation of cultural heritage, and the transparency of proprietary model development.
AI model training requires massive datasets to identify patterns, linguistic nuances, and factual information. While much of this data is scraped from the open web, tech companies increasingly turn to physical media to access high-quality, long-form content that may not be available in digital repositories. At Groundwork, our analysis shows that the process often involves the 'guillotine-style' removal of book spines to enable high-speed automatic document feeders (ADFs). By converting physical pages into machine-readable text via optical character recognition (OCR), companies can ingest vast amounts of copyrighted literature into their training sets without the traditional limitations of digital licensing or public domain access.
Evidence surfaced through the use of tracking technology—specifically, an AirTag planted in a rare book—confirmed that bulk orders of literature were being routed to specific AI training facilities. According to 404 Media, the tracking data led investigators to a warehouse in Las Vegas, identified as 'VGT3,' where books were systematically deconstructed. The presence of internal branding depicting a dinosaur devouring a book underscores a corporate culture that prioritizes the rapid acquisition of training data over the preservation of the physical artifacts. This finding serves as a primary data point in the ongoing debate over whether AI development constitutes 'fair use' under current copyright law.
Copyright law provides authors and publishers with the exclusive right to reproduce their work. When tech firms purchase books and physically destroy them to digitize the content, they are essentially creating a derivative digital copy. While companies often argue that the resulting training data is transformative, legal experts remain divided. The destruction of unique or rare copies adds a layer of ethical complexity, as it effectively removes the original artifact from the secondary market, potentially impacting the value of remaining copies and the provenance of the work. As of 2026, several high-profile lawsuits are challenging whether the ingestion of copyrighted works into LLMs without compensation violates existing intellectual property statutes.
Keep exploring
More from GroundworkTransparency in AI development is currently minimal, as most companies treat their training datasets as protected trade secrets. When consumers and authors are unaware of how their data—or the cultural artifacts they value—is being utilized, trust in the AI ecosystem erodes. At Groundwork, we emphasize that the lack of disclosure regarding source material creates a 'black box' effect. If a company refuses to confirm whether specific training methodologies involve the destruction of physical media, it complicates the ability of regulators and the public to hold these organizations accountable for their supply chain practices.
Alternatives to the destruction of physical media exist, though they are often more expensive or time-consuming for corporations. These include:
- Partnering with libraries and archives to license digital versions of books, which preserves the original physical copies.
- Utilizing existing digital repositories such as Project Gutenberg or the Internet Archive, provided the licensing terms are respected.
- Investing in advanced non-destructive scanning technologies, such as robotic book scanners that utilize cameras and infrared sensors to digitize pages without removing spines.
For authors, publishers, and consumers concerned about the future of intellectual property, the path forward involves rigorous advocacy for transparency. This includes supporting legislative efforts that require companies to disclose the sources of their training data and demanding that AI firms adhere to ethical acquisition standards. By tracking the flow of physical goods and maintaining pressure on tech conglomerates, the public can demand a higher standard of accountability in the pursuit of artificial intelligence advancement. Decisions regarding the future of knowledge and culture should not be made in the dark; they require a balanced approach that respects both innovation and the sanctity of the written word.
