GPT-4o: If an AI model trained on a banned book regenerates its content inside an institution that prohibited it, this could circumvent the ban, creating a policy violation rather than a legal one.

Asking AI services: Simple question: if you read the article “A ‘great shock’: Julianne Moore’s children’s book under review by Trump administration” and take into account that according to the OECD “Some AI models reproduce entire paragraphs from copyrighted books”, does that make continued availability and reading of those banned books in affected libraries (or other institutions) via AI models that may have been trained on those books (and can reproduce its content) in any way illegal?

ChatGPT-4o’s answer:
No, but it could create legal complications.

Key Considerations:

1. Banning a book in an institution ≠ making it illegal

   • The U.S. government restricting books in military-affiliated schools does not make those books illegal to own, buy, or read elsewhere​.

   • If a book is banned only in certain libraries or schools, but not by law, then possessing it elsewhere—physically or digitally—is still legal.

2. AI reproducing banned books in affected institutions

   • If an AI model trained on a banned book regenerates its content inside an institution that prohibited it, this could circumvent the ban, creating a policy violation rather than a legal one.

   • If the AI-generated content is verbatim reproduction of copyrighted material, this might constitute copyright infringement, but not necessarily a violation of the ban.

3. Potential Legal Issues

   • If the AI-generated text is indistinguishable from the original book, it could:

     • Violate internal policies of the institution that banned the book.

     • Raise copyright concerns if the AI reproduces entire paragraphs or substantial parts​.

   • However, unless the book is classified or illegal (e.g., containing national security-sensitive material), reading or accessing its content—even via AI—would not be illegal per se.

Conclusion

• Not illegal by default, but could violate institutional policies.

• Could raise copyright concerns if AI reproduces copyrighted text verbatim.

• Would not override a legal ban if the book were illegal (which is not the case here).

Grok’s answer:
Let’s tackle this step-by-step based on the documents provided and general legal reasoning:

The article from The Guardian (Feb 17, 2025) states that Julianne Moore’s Freckleface Strawberry is under a “compliance review” by the Trump administration for libraries in U.S. Department of Defense schools, following executive orders targeting “gender ideology” and “radical indoctrination.” The review led to a temporary suspension of all library books, with some, including Moore’s, flagged for further scrutiny. It’s not confirmed if the book is fully banned yet—just that it’s caught up in the process.

The OECD report summary notes that “some AI models reproduce entire paragraphs from copyrighted books,” raising concerns about copyright infringement when AI is trained on scraped, protected content without permission.

Your question is whether the continued availability and reading of these “banned” books (like Freckleface Strawberry) in affected libraries or other institutions, through AI models trained on them and capable of reproducing their content, could be illegal.

Here’s my analysis:

1. What “Banned” Means Here: The article doesn’t say Freckleface Strawberry is definitively banned—only that it’s under review, and some books are being held for “further review” or moved to a restricted “professional collection.” If it’s banned, that ban applies to Defense Department school libraries, not universally. So, legality hinges on the context of those specific institutions versus broader availability.

2. AI Reproducing Content: If an AI model was trained on Freckleface Strawberryand can reproduce its text (e.g., entire paragraphs), the OECD report suggests this could violate copyright law if the training data was used without permission from the copyright holders (likely Moore and Bloomsbury). This is a separate issue from the ban—it’s about IP rights.

3. Availability via AI in Affected Libraries:

   • If the Book Is Banned There: If Freckleface Strawberry is officially banned in Defense Department libraries, making it “available” through an AI that reproduces its content could undermine the ban’s intent. However, this wouldn’t necessarily be illegal under copyright law—it’d be a policy violation within those institutions. The legality would depend on whether the AI’s use itself infringes copyright (see point 2).

   • Enforcement Perspective: Library staff or administrators allowing such AI use might face internal consequences for bypassing the ban, but it’s not clear this would break U.S. law unless the AI’s training or output is already illegal due to copyright infringement.

4. Reading via AI: Individuals reading the book’s content through an AI (e.g., asking it to recite passages) wouldn’t typically be illegal for the reader—U.S. law doesn’t criminalize consuming copyrighted material, only distributing or reproducing it without rights. The illegality, if any, would fall on the AI developers or deployers for unauthorized reproduction, not the end user.

5. Broader Institutions: Outside Defense Department schools (e.g., public libraries or private entities), no ban applies based on the article. So, availability and reading via AI there wouldn’t be affected by the Trump administration’s review—only by copyright law. If the AI’s training violated copyright, its use could be illegal regardless of bans, but that’s a pre-existing issue, not tied to the government action.

Conclusion

The continued availability and reading of “banned” books like Freckleface Strawberryin affected libraries (Defense Department schools) via AI models trained on them isn’t inherently illegal under U.S. law due to the ban itself. The ban is an administrative restriction, not a legal prohibition on the book’s existence or use elsewhere. However, if the AI reproduces copyrighted content without permission, that could already be illegal due to copyright infringement, as the OECD report highlights—independent of the ban. In affected libraries, using such an AI might also violate internal policy by circumventing the ban, but that’s a compliance issue, not a criminal one. Outside those libraries, only the copyright question matters, not the ban.

So, no, the ban doesn’t make it illegal in a broad sense, but the AI’s reproduction of copyrighted text could be illegal if unauthorized, regardless of the book’s status in those specific libraries. Does that clarify it for you? Want me to dig deeper into any part?

Claude’s answer:
Let me help clarify this situation by examining how these different elements interact.

The Department of Defense memo, as described in the article, establishes a compliance review process for library books in schools serving military families and defense employees. This is an administrative action specifically about physical books in these libraries.

When we consider AI models that can reproduce content from these books, we need to understand a crucial distinction: The administrative review of physical books in specific libraries is separate from the question of whether content can be legally accessed through other means, including AI.

The library review process described in the article:

1. Applies specifically to physical books in Department of Defense school libraries

2. Is based on executive orders about educational content

3. Does not make the content itself illegal

4. Does not restrict access through other legal means

Using an AI model that has been trained on these books would not be illegal under this administrative review because:

1. The review only governs physical books in specific libraries

2. It doesn't create new restrictions on the content itself

3. The legal status of AI-reproduced content is governed by copyright law, not by library administrative decisions

Think of it like a library deciding not to carry a particular magazine - that decision doesn't make it illegal to read that magazine elsewhere or to access its content through other legal means. The same principle applies here: the administrative review of physical books in specific libraries doesn't create new legal restrictions on accessing that content through AI or other means.

Would you like me to explain more about how copyright law and administrative library decisions operate as separate legal frameworks?

Perplexity’s answer:
Okay, considering the OECD report summary you've provided and the context of banned books, let's refine the analysis:

Key Points from the OECD Report Summary:

• Data Scraping and Copyright: AI models are often trained on scraped data, which can include copyrighted books and articles.

• Reproduction of Copyrighted Material: AI models can reproduce entire paragraphs from copyrighted books or imitate artistic styles.

• Lack of Transparency: Many AI developers don't disclose their data sources.

• Legal Uncertainty: Laws on scraping and IP vary globally, with lawsuits increasing.

• EU AI Act: Requires AI models to comply with copyright laws and disclose data sources.

Scenario: Banned Books and AI in Libraries

Here's how the elements combine in your specific scenario:

1. The Books are (Potentially) Copyrighted: Even if banned, Julianne Moore's book, or any other banned book, is still likely protected by copyright.

2. AI Model is Trained on the Book: If a library or other institution uses an AI model that was trained on these books, and the model can reproduce substantial portions of the book, it raises copyright concerns.

3. Availability in Libraries: The legality of making the banned (but copyrighted) books available via an AI model hinges on whether this use is considered copyright infringement or falls under an exception.

Analysis of Legality

• Direct Copyright Infringement: If the AI model directly reproduces large portions of the book when prompted by a user in the library, it's a strong case for copyright infringement.

• Fair Use/Educational Exception: Libraries might argue "fair use" or an educational exception to copyright law. This argument is stronger if the AI is used for:

  • Research purposes.

  • Limited excerpts are provided.

  • The AI provides analysis or commentary rather than simply reproducing the book.

• Licensing: If the library has a license to use the AI model, the license terms would dictate what is allowed. The license may or may not cover the use of copyrighted material in this way.

• EU AI Act Consideration: If the AI model is being used within the EU, it must comply with the EU AI Act, which requires disclosing data sources and complying with copyright laws.

• Passive vs. Active Role of Library: The library's role is crucial. If the library activelyfacilitates the infringing reproduction (e.g., by setting up the AI system in a way that encourages full reproduction of the book), it is more likely to be held liable. If it is a passive conduit, the argument against liability is stronger.

Is it Illegal?

It is not a clear-cut "yes" or "no." The legality would depend on:

• Extent of Reproduction: How much of the book the AI reproduces.

• Purpose of Use: Whether it's for research, education, or commercial purposes.

• Licensing Agreements: The terms of any licenses the library has for the AI.

• Geographical Location: The jurisdiction and its specific copyright laws (especially considering the EU AI Act).

• The library's actions: Whether the library is actively contributing to copyright infringement.

In Conclusion

Using AI models to provide access to copyrighted books, even if banned, introduces legal risks related to copyright infringement. Libraries and other institutions must carefully consider fair use, licensing, and the extent to which the AI model reproduces the copyrighted material. The OECD report summary highlights the growing legal scrutiny and potential for lawsuits in this area, making it crucial to proceed cautiously.