Report from Courthouse News
In Brief – A federal judge in New York has denied OpenAI’s motion to dismiss a direct copyright infringement claim brought by a group of authors who allege the AI giant violated federal copyright law by using their copyrighted works to train their popular generative artificial intelligence chatbot. The ruling was focused on the question of whether Chat-GPT created infringing outputs that were substantially similar to plaintiffs copyrighted works. District Judge Sidney Stein wrote in his 18-page opinion that summaries of copyrighted books, such as plaintiff George R.R. Martin’s “A Song of Ice and Fire”, and detailed outlines for possible sequels generated by the chatbot, could be found to be “substantially similar” by a reasonable jury, in particular through the use of the original works’ setting, plot and characters that are protected by copyright. That was enough to deny OpenAI’s motion to dismiss. However, the judge took pains to note that the ruling was focused only on the issue of substantial similarity, and that “Nothing in this opinion is intended to suggest a view on whether the allegedly infringing outputs are protected as fair uses of the original works.”
Context – A series of lawsuits in US courts will be central to determining if the training of generative AI systems violated US copyright law. As this decision highlights, written outputs involving copyrighted characters and stories don’t need to be full on replicas of the original works to violate copyright law, as authors of fanfiction should know. Instead, the application of the “fair use” exception to copyright infringement is likely to be the more decisive legal question, and this decision did not try to address it. There were two more relevant decisions from US District judges in June, one defending training as a fair use, the other arguing against that claim. In the EU, the copyright section of the General-Purpose AI Code of Practice for the AI Act is a key development. In the UK and Australia, government proposals to make copyright law more conducive to AI training have received pushback from copyright industries.