AI art and copyright remain legally unsettled

A court in Delhi just told India’s Copyright Office to decide within eight weeks whether an image made by an artificial intelligence system can be registered as a copyrighted artwork. The petitioner is Stephen Thaler, and the work at issue is titled “A Recent Entrance to Paradise,” which he says was generated by his system DABUS, short for Device for the Autonomous Bootstrapping of Unified Sentience. (thehindu.com) That sounds narrow, but it goes to a basic copyright question: who is the “author” when a machine makes the expressive choices. Thaler’s argument in India is that the work should fit under rules for computer-generated works, with rights going to the person who caused the work to be created. (thehindu.com) The United States has been moving the other way. In March 2023, the United States Copyright Office said registration depends on human authorship, and applicants must disclose material generated by artificial intelligence if it is more than trivial. (copyright.gov) The office sharpened that position again in a 2025 report after reviewing more than 10,000 public comments. Its bottom line was that using artificial intelligence as a tool can still leave room for copyright, but material generated entirely by a machine does not qualify on its own. (copyright.gov) So a designer using image generators is now in a split-screen world. One legal system is being asked whether a fully artificial intelligence-made image can be registered at all, while another says the safest claim is in the human selection, editing, arrangement, or other human-made parts around the output. (thehindu.com) (copyright.gov) At the same time, the lawsuits over how these systems were built are getting messier, not clearer. In one federal copyright case against OpenAI, a judge criticized a key company witness for relying on “hazy recollections” about Project Giraffe, an internal effort tied to reducing copyright infringement risk, and said OpenAI’s lawyer objected at least 200 times during the deposition. (mercurynews.com) That matters because copyright fights over artificial intelligence are not just about the final image, song, or paragraph on your screen. They are also about training data, internal safeguards, and whether companies can explain in court what they built, what they filtered, and what they knew. (mercurynews.com) Now lawmakers are entering from a third direction: liability. Wired reported on April 10 that OpenAI backed an Illinois bill that would limit when frontier model companies can be sued for “critical harm,” even in cases involving mass casualties or damages above $500 million, so long as the developer did not intentionally or recklessly cause the incident and published safety, security, and transparency reports. (wired.com) Put those three developments together and the rulebook starts to look unfinished in three separate places at once. One court is asking whether machine-made art can be owned, another case is testing how well an artificial intelligence company can defend its copyright practices under oath, and a state bill would narrow who pays when a powerful model causes catastrophic damage. (thehindu.com) (mercurynews.com) (wired.com) For anyone making commercial art with these tools, the practical move is boring paperwork. Save the prompt, save the draft sequence, save the layers, save the edit history, and be ready to point to the exact human decisions that changed the result from raw machine output into a finished work. (copyright.gov) That kind of provenance will not settle the law in Delhi, New York, or Illinois. But in April 2026, it is the closest thing creators have to a receipt for authorship while courts and lawmakers are still deciding what artificial intelligence is allowed to own, copy, or break. (thehindu.com) (mercurynews.com) (wired.com)