The United States Supreme Court ruled six to three on Monday that training artificial intelligence models on copyrighted works constitutes fair use under existing US copyright law, delivering a landmark decision that resolves one of the most consequential and closely watched intellectual property disputes in decades. The ruling, written by Chief Justice John Roberts, sides with AI developers who have argued that ingesting copyrighted text, images, and other content for the purpose of training machine learning systems is sufficiently significant to qualify for the fair use exception to copyright infringement.

The case, consolidated from multiple district court cases involving major AI developers and coalitions of authors, visual artists, musicians, and news publishers, had been closely watched by the technology and creative industries as a potential turning point that would either validate the business model underlying the AI industry or expose companies to potentially catastrophic copyright liability for content used in training large language models and image generators. The decision, coming after more than two years of litigation that generated an unprecedented volume of amicus briefs from across the technology, entertainment, publishing, and academic sectors, provides the legal clarity that AI companies have sought since the first training data lawsuits were filed in 2022.

Chief Justice Roberts’s majority opinion applied the four-factor fair use test established in previous Supreme Court precedents and found that AI training weighs in favor of fair use on three of the four factors. The opinion emphasized that AI training represents a ‘highly significant’ use of copyrighted works because the purpose of the training process – producing a statistical model capable of generating new outputs – is fundamentally different from the purpose of the original works. The court analogized AI training to the digitization and indexing of books in Authors Guild v. Google, noting that neither use enables direct consumption of the copyrighted content in its original form. Reuters Legal reported that the majority opinion explicitly declined to address whether the outputs of AI systems trained on copyrighted works could themselves infringe copyright, leaving that question for future litigation.

The three dissenting justices, in an opinion authored by Justice Sonia Sotomayor, argued that the majority had applied the fair use doctrine in a manner that effectively eliminates meaningful copyright protection for the creative works that AI systems consume at scale. “The majority’s decision today tells every creator in America that their life’s work may be harvested without compensation to train machines that will compete directly with them in the marketplace,’ Sotomayor wrote. ‘This is not what Congress intended when it enacted the Copyright Act, and it is not what the fair use doctrine was designed to accommodate.” The dissent called on Congress to enact specific legislation addressing AI training, noting that the statutory framework governing copyright was not designed with the present technological reality in mind. SCOTUSblog described the dissent as ‘unusually pointed in its criticism’ of the majority’s reasoning.

The creative industries reacted with sharp criticism to the ruling. The Authors Guild, the American Society of Journalists and Authors, the Screen Actors Guild, and multiple news publisher trade associations issued statements condemning the decision as a blow to creators and calling on Congress to legislate in the area. Several legislative proposals have been introduced in Congress that would require AI companies to obtain licenses or pay royalties for copyrighted training data, and advocates for the creative sector said the Supreme Court ruling would intensify their lobbying efforts for federal legislation. Law360 reported that legislation establishing a compulsory licensing regime for AI training data has gained bipartisan support in the Senate Judiciary Committee and could advance to the floor later in 2026.

For AI companies, the decision removes the most significant near-term legal risk facing the industry and validates the current practice of training on internet-scale data without per-work licensing. Shares of publicly traded AI and technology companies rose broadly following the announcement. However, legal analysts cautioned that the decision does not resolve all copyright questions surrounding AI, including the output infringement question the majority explicitly left open and the separate question of whether AI systems that memorize and reproduce specific passages of training data are engaging in direct infringement rather than fair use. Ongoing litigation addressing these narrower questions is expected to continue through the federal court system in the years ahead.