吃掉书的AI 以及翻书的那只手

This article exposes Project Panama, a secret initiative by AI firm Anthropic, and walks through the origins of the event and the reasons behind its widespread controversy. Key takeaways are as follows:

1. Basic event background: Launched in early 2024 with a budget of tens of millions of dollars, Project Panama sees Anthropic purchase millions of physical books published before 2022 in bulk through secondhand booksellers, then scan and destroy all copies. To date, approximately 2 million books have been destroyed. After the project was exposed, it garnered over 2 million views on X (formerly Twitter), with nearly 4,000 critical comments.

2. Core points of controversy: Anthropic’s core goal is to obtain training data for its large language model. By destroying physical copies after scanning, the company exploits a loophole in U.S. law to claim "fair use" of the content. However, the practice damages cultural heritage: most of the destroyed books are rare out-of-print titles with extremely limited circulation. A comparison with Google’s early book-scanning project highlights the erosion of humanistic values amid AI development.

This incident holds multiple key insights for brands operating in AI and cultural content sectors. Key takeaways are as follows:

1. Copyright compliance risk warning: AI copyright disputes have become a pervasive global legal issue. As of May 2026, 112 AI copyright-related lawsuits are pending worldwide, with nearly 80% concentrated in the U.S. Anthropic itself previously paid a $1.5 billion settlement for using pirated e-books to train its models. Brands developing AI businesses must prioritize copyright compliance in advance to avoid the risk of catastrophic compensation.

2. Brand reputation warning: The widespread public outcry over Anthropic’s secret book destruction shows that the public values cultural heritage deeply. Brands pursuing related projects cannot prioritize commercial interests alone; they must account for public cultural sentiment, or risk irreversible damage to their brand reputation.

3. Brand building takeaway: The public recognizes humanistic values in digital transformation. Brands can leverage these values to build goodwill and earn public acceptance.

This incident brings clear opportunities and risk warnings for secondhand booksellers. Key takeaways are as follows:

1. Market opportunity: Large AI model developers have massive bulk purchasing demand for long-tail niche physical books, which typically see extremely low circulation in the secondhand market. Large-volume orders can bring direct economic gains for sellers. Sellers can proactively aggregate scarce resources such as out-of-print academic monographs, local historical archives, and niche technical manuals to match large model firms’ purchasing needs, and open up new revenue streams.

2. Risk warning: The current end use of purchased books—scanning followed by destruction—has already sparked massive public controversy over damage to cultural heritage, and targeted regulation is likely to be introduced in the future. Sellers should closely monitor policy developments, balance commercial gains with industry ethics, and mitigate reputational and regulatory risks.

This incident reveals new commercial opportunities and development insights for book processing and digitalization factories. Key takeaways are as follows:

1. New commercial opportunity: Large AI model firms have strong demand for bulk digitalization of physical books. The full process includes multiple processing steps: hydraulic spine cutting, high-speed scanning, shredding and destruction, etc. Projects with tens of millions of dollars in budget and millions of books to process have already emerged. Relevant factories can develop dedicated business lines to serve large model clients and unlock new growth.

2. Upgrade insight: Unlike traditional library scanning, which requires manual handling of rare old books, this business prioritizes automated processing efficiency and eliminates the need for manual page turning. Factories can upgrade to automated assembly lines to improve large-batch processing capacity and adapt to client requirements.

3. Risk warning: This business is tied to significant cultural controversy. Factories must verify the compliance of orders before accepting them, track future policy changes, and mitigate potential risks.

This incident exposes core pain points and new growth trends in the large AI model industry, offering valuable insights for digital and AI service providers. Key takeaways are as follows:

1. Industry pain point: Large model developers have extremely strong demand for high-quality scarce training data, but copyright compliance remains a persistent industry bottleneck. Traditional licensing models are costly and inefficient, and cannot meet large models’ demand for massive training corpora. The industry urgently needs new compliant solutions.

2. Development trend: Large model firms have already begun bypassing existing online copyright frameworks by obtaining legally compliant data through offline purchases of physical books, followed by scanning and destruction. Demand for bulk digitalization of physical books will continue to grow, opening up considerable market space.

3. Strategic business direction: Service providers can develop targeted automated physical book scanning and processing solutions, or explore legal and compliant training data licensing and trading models to solve industry pain points and unlock new business growth.

This incident offers multiple insights on risk mitigation and business development for platform companies pursuing AI-related business. Key takeaways are as follows:

1. Risk mitigation: Copyright risk is the biggest legal threat to current AI platforms. More than 80% of global AI copyright lawsuits are concentrated in the U.S., and one firm has already paid a $1.5 billion settlement. Platforms developing AI training businesses must build out copyright compliance frameworks in advance to avoid costly compensation lawsuits.

2. Business opportunity: The massive demand for scarce training data from large model firms has created new trading scenarios. Platforms can build matching channels to connect large model developers, secondhand booksellers and processing factories, offer compliant data matching services, and open up new revenue growth.

3. Reputational risk mitigation: The public holds strong negative sentiment toward AI practices that damage cultural heritage. Platforms pursuing related business must balance commercial interests with public cultural interests, maintain transparent public communication, and avoid triggering public relations crises that damage platform reputation.

The exposed Project Panama represents a new development in the large AI model industry, and carries important value for industrial research. Key takeaways are as follows:

1. New industry trend: Demand for training data has expanded from online pirated e-books to offline physical books. Leading industry players have developed a new model to avoid copyright issues by purchasing physical books, scanning them, then destroying the original copies. This is a new change as the AI industry enters the large-scale training phase.

2. New exposed problems: While this model complies with existing law, it causes irreversible damage to cultural heritage: a large number of rare, out-of-print niche titles have been permanently destroyed, undermining cultural diversity. It also highlights the mismatch between existing copyright law and the pace of AI industry development.

3. Research insight: Researchers need to prioritize ethical and legal issues in AI development, push for policy improvements, balance the interests of AI industry growth, copyright holders and cultural heritage preservation, and explore reasonable paths to preserve humanistic values amid AI development.