New York Times reporter and Bad Blood author John Carreyrou has joined five other writers in suing Google, OpenAI, Meta Platforms, Elon Musk’s xAI, Anthropic, and Perplexity, claiming their large language models were trained using copyrighted works without authorisation.
Filed in a California federal court on 22 December 2025, the lawsuit accuses the defendants of “wholesale copying” and ingestion of protected literary works to train chatbots and text-generation systems now deployed across consumer and enterprise applications.
Unlike previous author-led class actions, Carreyrou and his co-plaintiffs are pursuing individual claims. Their legal team argues that collective settlements have yielded “negligible recovery” for authors while allowing technology companies to continue profiting from what the filing describes as “uncompensated creative labour.”
The complaint follows a growing pattern of copyright disputes targeting AI developers. Earlier this year, Anthropic reached a $1.5 billion settlement in a separate case, though individual writers reportedly received only nominal sums. This latest filing signals a shift in strategy — from group restitution to tailored litigation aimed at testing how far copyright law can reach into the neural networks behind generative AI.
Carreyrou’s case also widens the field of defendants. While OpenAI and Meta have faced multiple suits from creators and publishers, this is among the first to include xAI, Elon Musk’s fast-growing AI business, which recently launched its Grok chatbot. Google and Perplexity are named for allegedly ingesting the same material through data-scraping pipelines.
Industry observers expect the defendants to argue fair-use exemptions under US copyright law, a defence that has featured prominently in prior AI cases. Courts have yet to define whether training a model on copyrighted text — without reproducing it directly — constitutes infringement.
Legal specialists describe the move as both symbolic and strategic. “This is about setting precedent,” said one copyright lawyer not connected with the suit. “By bypassing class certification, these authors can press for maximum statutory damages and keep their claims intact if future rulings redefine fair use in the AI context.”
The outcome could have material consequences for how AI businesses source and license data. Several governments, including India and the EU, are already weighing proposals for royalties tied to generative-model training, reflecting a global reassessment of digital-content value chains.
For technology companies, the case underscores the operational risk of opaque data practices. For creators and publishers, it marks a calculated escalation — testing whether individual authors can compel an industry that has, until now, treated training data as public domain.
You must be logged in to post a comment.