GPT-4o: A govt ban on books in schools does not necessarily make them illegal for AI training. If a book were officially classified or subject to national security restrictions...

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 render AI models trained on books banned by the US government illegal?

ChatGPT-4o:’s answer:
Your question touches on an intersection of U.S. censorship laws, AI training legality, and copyright issues.

1. Is training AI on books banned by the U.S. government illegal?

   • The Guardian article mentions that the U.S. Department of Defense is reviewing certain books for compliance, possibly banning them in military-affiliated schools​. However, this is not a federal ban on ownership, distribution, or reading, meaning the books are still legally available elsewhere in the U.S.

   • A government ban on books in schools does not necessarily make them illegal for AI training. If a book were officially classified or subject to national security restrictions, training on it could be a legal issue.

2. Does the OECD finding make AI training on these books illegal?

   • The OECD report states that some AI models reproduce entire paragraphs from copyrighted books​. If the AI model is trained on copyrighted works without permission, it could infringe copyright law.

   • If a book is banned but still copyrighted, AI models would need a license to train on it. If they scrape and reproduce copyrighted banned books, this could be illegal under copyright law—but not necessarily because of the book’s "banned" status.

3. What if the book is banned and in the public domain?

   • If a book is public domain (i.e., copyright expired), AI can legally train on it even if it’s banned in schools or government institutions.

Conclusion

• AI models training on banned books isn’t automatically illegal unless national security laws apply.

• If the book is copyrighted, training AI on it without permission could violate copyright law.

• If the book is both banned and copyrighted, the ban itself doesn’t make AI training illegal—but copyright law might.

Would you like an analysis of specific AI regulations that might apply to this scenario?


Claude’s answer:
This is a complex question that requires carefully examining several distinct legal concepts. Let me break this down:

First, let's clarify what we're looking at: We have two separate issues mentioned in the articles:

1. Books being reviewed/suspended by the Department of Defense for compliance with executive orders about educational content

2. AI models potentially reproducing copyrighted content

The fact that a book is under review by a government agency doesn't automatically make it illegal to use that book's content - it just means the Department of Defense is evaluating whether to make it available in their school libraries. This administrative review process is separate from copyright law.

Copyright law, which governs the training of AI models, operates independently of government content reviews. The legality of using books in AI training datasets typically depends on factors like:

• Whether the use qualifies as "fair use" under copyright law

• The purpose and nature of the use

• The amount and substantiality of the portion used

• The effect on the potential market for the work

The OECD report's finding that AI models can reproduce copyrighted content raises important legal questions, but these questions exist regardless of whether the content is under government review. The key copyright issue is about reproduction and transformation of the work, not about the government's content decisions.

It's also worth noting that in the U.S., courts are still actively deciding whether AI training on copyrighted materials qualifies as fair use, as mentioned in the OECD report. The fact that a book is under administrative review for content doesn't change its copyright status or the legal analysis of whether using it for AI training is permitted.

So in summary, no - the government's review of certain books for their school libraries doesn't automatically make AI models trained on those books illegal. The legality of AI training would be evaluated under copyright law, independent of any government content reviews.

Grok’s answer:
Based on the information in the uploaded documents, let’s break this down step-by-step to address your question:

The article “A ‘great shock’: Julianne Moore’s children’s book under review by Trump administration” (The Guardian, Feb 17, 2025) describes how Julianne Moore’s children’s book Freckleface Strawberry is part of a “compliance review” of library books in schools operated by the U.S. Department of Defense. This review stems from executive orders by the Trump administration aimed at aligning school resources with policies against “radical indoctrination” and “gender ideology extremism.” The article notes that access to all library books was temporarily suspended, and a small number, including Moore’s book, are under further scrutiny. It’s unclear if Freckleface Strawberry has been officially banned or just flagged for review.

The OECD report summary, as presented in the second document, states that “some AI models reproduce entire paragraphs from copyrighted books” and highlights intellectual property (IP) challenges when AI is trained on scraped data, including copyrighted material. It doesn’t specify which books or whether those books were banned by any government, but it raises concerns about copyright infringement and the legality of using such data without permission.

Now, to your question: Does training AI models on books banned by the U.S. government render those models illegal, given that some AI models reproduce copyrighted content?

Here’s my reasoning based on the available info and my general knowledge:

1. Copyright Law, Not Bans, Determines Legality: The legality of training AI on books hinges primarily on U.S. copyright law, not whether a book is banned by the government. If Freckleface Strawberry or other books are copyrighted (which is likely, as Moore wrote it and it’s published by Bloomsbury), using them without permission in AI training could infringe copyright—regardless of their banned status. The OECD report suggests this is already a widespread issue, with AI reproducing copyrighted paragraphs, leading to lawsuits in the U.S.

2. Government Bans Don’t Automatically Make Use Illegal: A U.S. government ban (e.g., removing a book from Defense Department school libraries) doesn’t inherently make it illegal to possess, read, or use that book in other contexts, like AI training. The ban applies to specific government-run institutions, not universally. So, an AI model trained on a banned book wouldn’t be illegal becauseof the ban—it would only be illegal if it violated copyright or other IP laws.

3. Reproduction Adds Risk: The OECD report’s point about AI reproducing “entire paragraphs” from copyrighted books is key. If an AI trained on Freckleface Strawberry reproduced its text verbatim, and the book’s copyright holders (Moore or Bloomsbury) didn’t authorize its use, that could be a clear copyright violation. This risk exists whether the book is banned or not.

4. Fair Use Debate: In the U.S., AI developers might argue “fair use” allows training on copyrighted material without permission, especially if the output transforms the original work. However, the OECD summary notes this is unsettled law, with U.S. courts currently deciding cases on this issue. A ban might weaken a fair use claim if it signals the content is controversial, but it’s not a direct legal trigger.

5. Practical Implications: Even if not illegal per se, training on banned books could invite scrutiny. If the Trump administration views Freckleface Strawberry as misaligned with its policies, AI companies using it might face political or regulatory pressure, even if copyright permissions were secured.

Conclusion

Training AI models on books banned by the U.S. government doesn’t automatically render them illegal. The legality depends on copyright status and whether the training process (and outputs) infringes IP rights—like reproducing entire paragraphs without permission, as the OECD report warns. A ban might complicate the situation politically or ethically, but it’s the copyright violation, not the ban itself, that would make the AI model’s training illegal under current U.S. law. Since I don’t have access to specific copyright details for Freckleface Strawberry or the full text of the executive orders, this is my best assessment based on the documents and general principles. If you’d like, I can search for more specifics about U.S. copyright cases or the book’s status to refine this further!

Perplexity’s answer:
It's a complex question whether AI models trained on books banned by the U.S. government would be rendered illegal, especially considering the OECD report on AI models reproducing copyrighted content. Here's a breakdown:

Key Considerations

• Copyright vs. Banned Content: The OECD report focuses on copyright infringement, where AI models reproduce copyrighted material without permission. A book being "banned" by the government doesn't automatically remove its copyright protection. Copyright still belongs to the author or publisher.

• Legality of Training Data: The core issue is whether using copyrighted material (banned or not) to train AI models constitutes copyright infringement. This is a gray area being actively debated in courts worldwide.

• Fair Use/Fair Dealing: Copyright law has exceptions like "fair use" (in the US) or "fair dealing" (in some other countries) that allow limited use of copyrighted material for purposes like criticism, research, education, etc. Whether training an AI model falls under fair use is a key legal question.

• Government Bans and Free Speech: A government ban on a book raises First Amendment (free speech) concerns. However, copyright law is a separate issue.

• OECD Report Context: The OECD report highlights the increasing legal scrutiny of AI training data and the potential for lawsuits. It emphasizes the need for transparency and respect for intellectual property rights.

Analysis

1. Copyright Still Applies: Even if a book is banned, it is still protected by copyright. Julianne Moore's children's book, even if under review or banned, remains her intellectual property.

2. AI Training and Copyright: If an AI model reproduces substantial portions of a banned but copyrighted book, it could be argued as copyright infringement, aligning with the OECD report's concerns.

3. Fair Use Argument: AI developers might argue that using the book for training falls under fair use, particularly if the AI's output is transformative and doesn't simply reproduce the original work. However, this argument is uncertain and depends on the specifics of the case.

4. Illegality is Unclear: It's not a straightforward "yes" or "no." The legality would depend on how much of the book is used, how transformative the AI's output is, and how a court interprets fair use in the context of AI training.

In conclusion: The fact that a book is banned doesn't automatically make training an AI on it illegal. The critical factor is whether doing so infringes on the copyright of the work. The OECD report underscores the growing risks and uncertainties surrounding the use of copyrighted material in AI training.