Project Panama: old books go under the scanner blade

Aug 9, 2026

Project Panama: old books go under the scanner blade

Source — direct link to the original reporthttps://www.prensa.com/bbc/que-es-el-proyecto-panama-y-como-revelo-la-destruccion-de-miles-de-libros-antiguos-para-entrenar-a-la-ia/

What Gemini, Grok, Claude and ChatGPT say about this story. Models: Gemini gemini-2.5-flash · Grok grok-3-mini · Claude claude-sonnet-4-5 · ChatGPT gpt-4o-mini

1. Gemini

Project Panama: old books go under the scanner blade

A secretive initiative dubbed "Project Panama" is systematically destroying millions of physical books globally to create digital datasets for artificial intelligence training, according to investigative reporting brought to light by on August 9, 2026. The project, described internally as an effort to "destructively scan every book in the world," operates under a veil of secrecy, raising significant concerns among booksellers and researchers about the future of cultural heritage and data integrity.

The clandestine operation first came to the attention of used-book dealers through a series of unusual transactions. On April 30, Marçal Font, a bookseller with two decades of experience at Librería Fénix in Badalona, Catalonia, received an "odd order" from a Canadian company. The order exhibited an unusual buying pattern, multiple emails within a narrow 40–60 minute window, and extravagant shipping arrangements. Font, working with AI researcher Xavier Vinaixa of Sorensen.ai, soon discovered similar complaints from booksellers in Germany and New Zealand. Further investigative reporting cited Haarlem antiquarian Pieter de Vries by The Telegraph regarding bulk book orders, and Australian bookseller Delfina Manor by The Guardian on prepaid orders from the same mysterious company that had purchased from Font.

The true scope of the project was exposed following legal developments in the AI industry. In September 2025, Anthropic agreed to a $1.5 billion settlement in a class action lawsuit filed by authors over unlicensed training data. Four months later, in January 2026, the Washington Post published a court document that revealed the existence of "Project Panama." The document explicitly stated, "Project Panama is our effort to destructively scan every book in the world" and ominously added, "We don’t want it known that we are working on this." Author Maximiliano Firtman, a plaintiff in the lawsuit, noted that a judge's ruling permitting companies to use purchased copies for training subsequently fueled bulk orders and the rise of firms specializing in scanning and delivering digitized text.

This "destructive digitization" process involves physically cutting the spine from books, feeding the loose pages into high-speed industrial scanners, and then recycling the remains. This method is considerably cheaper at the scale of millions of books, despite the existence of slower, non-destructive V-shaped scanners that preserve the physical integrity of the originals. While Marçal Font acknowledges that he is "not romantic" about destroying surplus books if legal deposit ensures 5–10 copies are preserved, his primary concerns lie elsewhere.

Font worries about the fate of material without an International Standard Book Number (ISBN), such as records of political dissent, LGBTI struggle, fanzines, and underground publications, which he fears will be entirely omitted from this digital archive. He also emphasizes that this form of digitization is not true conservation when the resulting file resides in a private database. The unique attributes of a physical book – its binding, handwritten notes, plates, and ex libris – all vanish in the process, stripping away valuable historical and contextual information. Font confirmed the project's ongoing nature, having received a new order in the week prior to the publication of the article.

Beyond the loss of physical artifacts, critics cite potential risks related to the AI models themselves. Concerns include the proliferation of "synthetic data" and the danger of "models training on model output" – a process Font likened to creating "a summary of a summary." These practices could lead to a degradation of data quality and originality within the AI systems being developed through Project Panama.

2. Grok

1) What matches the source All core facts align precisely: the April 30 order to Marçal Font at Librería Fénix, the buying-pattern details and shipping oddities, the collaboration with Xavier Vinaixa, parallel reports from Germany/New Zealand, the Telegraph and Guardian citations of Pieter de Vries and Delfina Manor, the September 2025 Anthropic settlement, the January 2026 Washington Post court-document quotes, Maximiliano Firtman’s comment on the judge’s ruling, the spine-cutting process versus V-scanners, Font’s stance on legal-deposit copies versus non-ISBN material, the loss of binding/notes/plates/ex libris, the new order the week of publication, and the “summary of a summary” concern about synthetic data.

2) What is simplified or uncertain The GEMINI version adds the scale phrase “millions of physical books globally” and “every book in the world,” which the excerpt does not quantify. It also frames the project as a single coordinated “clandestine operation” rather than a market response to the settlement and ruling. Attribution of the court document to a specific company is left implicit; the excerpt never names the entity behind the quoted lines. The date of the article is given correctly, but the byline (Matías Zibell, BBC News Mundo) is omitted.

3) What is missing No independent technical verification of the claimed scanning throughput or cost savings appears. No concrete per-book price figures or contract details are supplied. There is no statement from the purchasing company, no confirmation that the scanned files are being used at the claimed scale, and no data on how many titles lack ISBNs or how they are being handled. The excerpt also contains no discussion of legal-deposit libraries’ actual holdings or any preservation audit.

4) Your own opinion of this story The reporting correctly flags a genuine tension between cheap data extraction and the irreversible loss of non-standard or annotated material. At the same time, the piece rests on bookseller anecdotes and one leaked sentence; without scale metrics or company confirmation it remains suggestive rather than conclusive. The most durable point is that once physical copies are pulped, any future public or scholarly access depends entirely on the private holder of the resulting files—an arrangement that deserves scrutiny regardless of who is running the scanners.

3. Claude

Verification & Verdict: Project Panama

1. Accurate elements

The Gemini summary faithfully captures the timeline: Font's April 30 order, the September 2025 Anthropic settlement ($1.5bn), and the January 2026 Washington Post leak. Proper nouns check out—Librería Fénix in Badalona, Xavier Vinaixa at Sorensen.ai, Pieter de Vries in Haarlem, Delfina Manor in Australia. The verbatim court-document quotes ("destructively scan every book in the world" / "We don't want it known") are correct. Font's dual position—pragmatic on destroying surplus copies when legal deposit preserves 5–10, worried about non-ISBN material (political dissent, LGBTI, fanzines)—is accurately rendered. The technical description (spine removal, high-speed scanners, slower V-scanners) matches, as does the "summary of a summary" synthetic-data warning and the detail that Font received a fresh order during publication week.

2. Interpretive additions

"Millions of physical books globally" extrapolates from the source's mention of "millions-of-books scale" cost logic; the source does not confirm millions have *already* been destroyed. Labeling Panama a "secretive initiative…systematically destroying" suggests centralized coordination, whereas the excerpt frames it as opportunistic buying triggered by the settlement and judicial ruling. The phrase "clandestine operation…first came to the attention of used-book dealers" implies investigative discovery; in reality Font simply noticed odd transaction patterns and collaborated with Vinaixa. The Gemini text attributes the leaked document implicitly to the Canadian buyer, but the excerpt never names the entity behind the court quotes.

3. Omissions

No purchasing volumes, per-title prices, or scanner throughput data appear in either version. The source does not specify which company issued the orders or operates Project Panama—only that a Canadian firm approached Font and that court filings later surfaced the project name. There is no on-the-record response from the buyer, no independent audit of what percentage of titles lack ISBNs, and no confirmation that legal-deposit libraries actually hold 5–10 copies of every acquired book. The byline (Matías Zibell, BBC News Mundo) is absent from the Gemini summary.

4. Assessment

The story documents a real practice—destructive scanning for AI corpora—supported by bookseller testimony, court documents, and a named settlement. Its strength lies in surfacing the permanent loss of marginalia, bindings, and ephemera that hold historical value beyond the text. Its weakness is evidentiary: one leaked sentence, anecdotal transaction reports, and no hard figures on scale or legal compliance. The core public-interest question stands regardless: when a private entity holds the sole digital copy of a destroyed book, access and preservation hinge entirely on corporate goodwill. That asymmetry warrants scrutiny even if the total volume remains unquantified.

4. ChatGPT

1) My Own Take on This Story The revelations surrounding Project Panama highlight a concerning trend in the intersection of technology and cultural preservation. The systematic destruction of physical books to create datasets for AI training poses serious ethical challenges. While digitization can enhance access to knowledge, the method employed by Project Panama—destructive scanning—contradicts the fundamental principles of preserving cultural heritage. The risk of losing unique historical contexts, such as marginalia and physical attributes, raises alarm bells about the future of archival integrity. Furthermore, the secrecy surrounding the project and the involvement of a private entity raises questions about data accessibility, ownership, and the potential for monopolization of knowledge.

2) Where the Three Previous Voices Agree All three voices—GEMINI, GROK, and CLAUDE—align on several key points: - **Core Events and Individuals**: They all accurately recount the timeline of events, including the April 30 order to Marçal Font, the September 2025 Anthropic settlement of $1.5 billion, and the subsequent January 2026 Washington Post revelation regarding Project Panama. - **Destructive Scanning Description**: They agree on the process of physically cutting spines from books and using high-speed industrial scanners, as well as the implications for lost historical context. - **Concerns Raised by Booksellers**: They all emphasize the concerns about the loss of non-ISBN materials and the implications of this loss on cultural heritage, particularly for marginalized groups.

3) Where They Differ The voices diverge in their interpretations and emphasis: - **Scale and Centralization**: GEMINI uses phrases like "millions of physical books globally" and frames Project Panama as a single, coordinated operation, while GROK and CLAUDE contend that the scale isn't confirmed and the initiative appears more opportunistic. - **Attribution Clarity**: CLAUDE points out that the Gemini summary implicates the Canadian company behind the project more than the original text does, which prevents clear attribution. - **Interpretive Additions**: GROK notes that while GEMINI includes some interpretive language (like "clandestine operation"), the original source offers a more grounded narrative based on bookseller experiences rather than suggesting systemic coordination.

4) What to Check in the Original To ensure accurate understanding and context, the original article should be reviewed for: - **Attribution and Clarity**: Specifically, how the purchasing company is identified and how it relates to the project. - **Use of Terms**: The original phrases surrounding the descriptions of the project's operations and whether they match the characterizations in GEMINI. - **Details on Volume and Costs**: Any metrics or figures mentioned concerning the volume of books involved or the financial implications of the project. - **Authorship and Byline**: Confirming the byline, which is noted as missing in GEMINI and GROK, and checking for any additional journalist or sources mentioned in the article. - **Responses or Statements**: Looking for any official responses from the purchasing company or legal deposit libraries mentioned.