ChatGPT starts blocking direct requests to copy authorial style

Journalists at Ars Technica tested ChatGPT by asking it to write an introduction to a story in the style of Stephen King, and received a refusal: the model stated that it could not accurately imitate the author's "distinctive voice," but offered a text with an atmosphere of horror and small-town anxiety in its own rendition. The model gave similar evasive responses to requests in the styles of J.K. Rowling, Amy Tan, Charles Dickens, and Ernest Hemingway. Earlier analysis from the publication No Latency showed that ChatGPT refused to copy the style of living authors but fulfilled such requests for deceased writers. This behavioral change could have legal significance amid numerous lawsuits from book authors accusing OpenAI of widespread copyright infringement in training models on their texts—one lawsuit directly points to ChatGPT's "supernatural ability" to generate text similar to copyrighted material. In the United States, copyright law protects the specific expression of an idea, not an author's abstract style; however, imitation may be deemed infringement if it becomes "substantially similar" to the original. Robert Brauneis, a law professor at George Washington University, told Bloomberg Law that never before has the personal style of individual creators been imitated as effectively and cheaply as AI does. A representative for OpenAI did not respond to Ars Technica's request for comment.