Can AI companies train on copyrighted work? What the lawsuits by newspapers, photographers and authors are about

Whether AI companies may train on copyrighted work is not settled; courts are deciding it case by case, and different judges have emphasised different things. Large language models are trained on data scraped from the Internet, often including copyrighted material. U.S. developers have traditionally believed this is fair use because it is transformative and limited. Authors, photographers and newspapers have sued, saying that AI companies should compensate creators.

The cases vary. The New York Times sued OpenAI and Microsoft in December 2023, alleging that their models were trained on millions of its articles without authorization. OpenAI says its use is fair use. In 2025 the judge denied most of OpenAI's motions to dismiss, which let most of the Times' claims go forward; that is a step in the case, not a verdict.

Getty Images sued Stability AI in London in 2023 for using its images in training data. As of November 4, 2025, Getty had largely lost that UK case. A separate US case continued, and in April 2026 a federal court let most of Getty's claims proceed.

The largest sum so far came in a case brought by authors. In Bartz v. Anthropic, a federal court ruled in June 2025 that training on digital copies was fair use, but ordered a trial over the acquisition and retention of pirated books. Anthropic then agreed to pay $1.5 billion to settle the case, and a judge approved the settlement in July 2026. So far, the answer depends on how the works were obtained and what harm was shown.