Are Generative AI Models Built on Stolen Creative Work? Unpacking Two Court Verdicts
Currently, at least 21 lawsuits in the US target tech companies for training AI models using copyrighted material. Writers, music labels, and news agencies argue this practice amounts to theft. Two recent court rulings have favored AI developers, making these among the first decisions addressing whether generative AI models are built on stolen creative content.
At their core, AI models like ChatGPT and Gemini operate by learning patterns from vast datasets. Their output—be it text, images, or music—depends heavily on the quality and scope of the data used during training. This data often includes books, articles, images, sounds, and other content freely available on the Internet.
Tech companies defend their approach by claiming their AI models create “transformative” works. This falls under the legal concept of “fair use,” which allows limited use of copyrighted material for public benefit, such as quoting excerpts in reviews. The following outlines the two key cases and explains why their outcomes matter.
Case 1: Writers vs Anthropic
In August 2024, a group of journalist-writers filed a class action lawsuit against Anthropic, the company behind the Claude family of Large Language Models (LLMs). They claimed Anthropic downloaded pirated copies of their works and used them without compensation to train its AI models. The concern was that such use undermines authors' ability to earn a living, as AI-generated texts could replace paid writing.
Anthropic reportedly used Books3, a shadow library with around seven million pirated books, for training, while also investing millions in purchasing and scanning printed books to build a general research library.
On June 23, Judge William Alsup ruled that Anthropic’s use of copyrighted material qualified as “fair use.” He emphasized the transformative nature of AI, stating that the models do not simply replicate original works but create something new and different. If copying is necessary during training, it falls under transformative use.
Case 2: Writers vs Meta
Thirteen authors, including comedian Sarah Silverman and writer Ta-Nehisi Coates, sued Meta over its Llama LLMs. They argued Meta copied massive amounts of their text, using it to train models that output content derived from their work. Meta also sourced data from Books3, Anna’s Archive, and Libgen—other shadow libraries.
Meta defended itself by explaining it “post-trained” its models to prevent memorization or direct output of copyrighted texts. They claimed their safeguards limited the AI from generating more than 50 words verbatim from any specific copyrighted source.
On June 25, Judge Vince Chhabria ruled that the plaintiffs failed to prove market harm. He explained that if AI-generated biographies flooded the market, it could dilute demand for original works—but this has not happened. Though acknowledging AI’s innovative role, the judge suggested that tech companies profiting from AI should find ways to compensate creators.
Why These Rulings Matter
Both rulings are wins for Anthropic and Meta but don’t grant them complete immunity. Questions remain about the legality of downloading content from pirated databases. In Anthropic’s case, the court found that storing and copying pirated books infringed copyright and scheduled a trial to determine damages.
Meta will also face further scrutiny on piracy issues. Anthropic is dealing with separate lawsuits from music publishers over training its AI on copyrighted lyrics. Many similar cases are underway, including a consolidated lawsuit against OpenAI and Microsoft involving authors, newspapers, and publishers.
Visual artists have filed suits against image-generating AI tools like Stability AI and Midjourney for using their work without permission. Getty Images sued Stability AI for using millions of copyrighted photos. In India, news agency ANI and several media organizations have taken legal action against OpenAI for unauthorized use of Indian copyrighted content.
These cases highlight that while courts have provided some clarity, the question of AI and copyright is far from settled. As AI-generated content grows, the creative community must consider the impact on livelihoods and the future of originality.
For creatives looking to understand AI’s evolving role and legal landscape, staying informed is crucial. Exploring resources like Complete AI Training’s latest courses can offer valuable insights into AI tools and responsible use.
Your membership also unlocks:
