Courts are deciding whether developers may use copyrighted books, journalism, images and other works when training generative AI. On September 2, the federal government entered the New York Times litigation against OpenAI and argued that AI training can qualify as fair use, emphasizing economic, scientific and national-security interests. Publishers and other creators argue that valuable copyrighted work should not become uncompensated raw material for commercial AI systems. The dispute reaches beyond one lawsuit because different rules could affect licensing markets, creator income, model development costs and the competitiveness of the U.S. AI industry. AI can learn from culture without receiving a blank check to consume it. A workable rule should protect experimentation while recognizing that professional creators also need a functioning market for their work.
Where should federal copyright law draw the line on using protected works to train AI?