Suchir Balaji, a researcher who helped gather and organize internet data used to train GPT-4, left OpenAI in August 2024 after concluding that its business practices violated copyright law. In an essay published in October 2024 and an October 23 interview with The New York Times, he argued that systems such as ChatGPT were trained on competitors’ and creators’ works and could generate outputs that imitate or substitute for those works. He contended that the systems failed the four-factor U.S. fair-use test, while OpenAI maintained that its models relied on publicly available data and were protected by fair use and related international principles.
The claim remained legally contested rather than an established finding. The New York Times’ attorneys identified Balaji in a November 18, 2024 court filing as a person who might possess relevant documents in its copyright lawsuit against OpenAI, and he said he would testify, but the source states that he had not disclosed previously unknown inside information. Stanford professor Mark Lemley disputed his legal conclusion, and attorneys later described the essay as an intelligent but incomplete non-lawyer analysis. Balaji died on November 26, 2024; after public speculation and claims by his parents of foul play, the San Francisco Police Department and Chief Medical Examiner concluded there was no evidence of homicide and ruled the death a suicide in an autopsy released February 14, 2025. The copyright lawsuits and regulatory debate over AI training data continued.
