Five publishers including Scott Turow of the Authors Guild have filed suit against Meta and Mark Zuckerberg over alleged copyright violations in AI training. The case joins a growing wave of litigation testing whether AI companies need to license training data.
The AI training data wars are heating up. OpenAI has faced multiple lawsuits from authors and newspapers. GitHub has fought claims over Copilot's use of public code. Now Meta—which has been relatively quiet in these fights—faces its own reckoning over what its models trained on.
The lawsuit, reported by The New York Times, alleges Meta used copyrighted books to train its Llama language models without permission or compensation. The publishers argue this constitutes mass copyright infringement on an unprecedented scale.
Meta's defense—and the industry's broader argument—rests on fair use. Training AI models on existing text is transformative use, the companies claim. The output isn't copying the input; it's learning patterns from it. No different than a human reading books to understand how language works.
The publishers aren't buying it. When Meta trains on millions of books and then offers AI services that can summarize, analyze, or create content in similar styles, that's not transformative use—it's industrial-scale substitution.
What makes this case potentially more significant than previous attempts is the evidence trail. Meta has been more open than OpenAI about its training process and model architecture. That transparency might create stronger legal footing for copyright claims.
The litigation reflects a fundamental tension in AI development. These models need vast amounts of text to work effectively. If every piece of training data requires licensing, AI development becomes prohibitively expensive—and dominated by whoever has the deepest pockets.
But if AI companies can train on anything without permission, they're building billion-dollar businesses on the unpaid labor of creators.
