OpenAI is facing renewed criticism over claims that it used copyrighted material without authorization to train its AI models. A fresh report from the AI Disclosures Project suggests that the company may have relied on non-public books to develop GPT-4o, its most advanced AI model to date.

How AI Training Works and Why Copyright Matters

AI models like GPT-4o are built using vast amounts of data, including books, articles, and multimedia sources. These models don’t “think” like humans; instead, they generate text based on patterns in their training data.

The controversy lies in the origins of this data. If AI firms are using copyrighted books without permission, they may be violating intellectual property laws, sparking both ethical and legal debates.

Report Alleges OpenAI Trained on O’Reilly Media’s Paywalled Books

The AI Disclosures Project claims that GPT-4o appears to recognize content from books published by O’Reilly Media—books that were never made publicly accessible.

Using a method called DE-COP, which analyzes AI-generated responses against original texts, researchers found that:

• GPT-4o displayed a strong familiarity with O’Reilly books published before its last known training date.
• Its predecessor, GPT-3.5 Turbo, showed significantly less recognition of paywalled books, implying that OpenAI may have introduced new training data.
• Even after accounting for general AI performance improvements, the findings suggested potential reliance on copyrighted material.

Does This Prove Copyright Infringement?

While the study’s findings raise concerns, the authors acknowledge that their method isn’t conclusive. It’s possible that OpenAI didn’t directly train on O’Reilly books but instead learned from snippets submitted by ChatGPT users.

Additionally, the report does not assess OpenAI’s newest models, such as GPT-4.5 or specialized reasoning models like o3-mini and o1. This means there is no definitive proof that OpenAI is still incorporating paywalled books into its training data.

The Bigger Picture: AI Firms Compete for High-Quality Data

OpenAI isn’t the only company under scrutiny for its data collection practices. As AI models advance, the race for high-quality training data has intensified. Many companies have taken steps to refine their AI training strategies, including:

• Hiring journalists and industry experts to fine-tune AI-generated content.
• Paying for access to news outlets, social media data, and stock media libraries.
• Introducing opt-out options for copyright holders—though these systems have been criticized for their limitations.

Despite these measures, OpenAI has continued to advocate for more flexible copyright regulations in AI training. This stance has already led to multiple legal battles.

What’s Next for OpenAI?

As lawsuits and regulatory scrutiny mount, OpenAI has yet to formally address the latest allegations. Whether these claims will lead to legal action remains uncertain, but one thing is clear—AI training practices and data ethics will remain a heated topic in the tech industry.

Would you trust AI models trained on copyrighted content? Share your thoughts!

Get the Latest AI News on AI Content Minds Blog