Justice Department says AI training can qualify as fair use

The US Justice Department has told a Manhattan federal court that training large language models on copyrighted written works can qualify as fair use, backing OpenAI and Microsoft in The New York Times' copyright suit. In a statement of interest filed on 1 September, the department said a contrary reading of copyright law would suppress innovation, weaken American competitiveness and compromise national security. The filing entered a case brought in 2023 in which The New York Times and other media organizations say OpenAI and Microsoft used millions of the paper's copyrighted news articles without permission to train systems including ChatGPT. The complaint seeks monetary damages, an order barring training on the paper's material and the destruction of any models trained on its work, while OpenAI and Microsoft do not dispute that their models were trained on the content and say the use was lawful. The Justice Department said that, consistent with national interests in innovation and free expression, training AI models on copyrighted material does not itself violate copyright law. It argued that excluding LLM training from fair use would be inconsistent with copyright principles and would severely hamper the progress of science and useful arts. The department also said the creative possibilities and public benefits of LLM training outweigh any competitive harm. In its filing, it portrayed licensing fees as a barrier that would mainly reward legacy media companies and leave only the biggest technology groups able to pay. National security featured prominently in the brief. The department cited a January 2025 executive order on American leadership in AI and a June order on encouraging innovation and responsible adoption, and it quoted the Government Accountability Office warning that failing to adopt and integrate AI could hinder national security, from intelligence analysis to battlefield decision-making. It said rules that make it significantly harder to build a robust US AI industry would hand an advantage to foreign adversaries. The filing also said the United States has a sovereign interest in the proper interpretation of federal law, including the Constitution's Intellectual Property Clause and the Copyright Act of 1976. It described LLMs as systems trained on billions of written works that predict the most likely next string of text from a prompt. The Copyright Office said in a report that dozens of lawsuits are pending in the United States on fair use and AI training, after it received more than 10,000 comments to a 2023 notice of inquiry.