Hours before the Register of Copyrights, Shira Perlmutter, was unceremoniously fired, the U.S. Copyright Office published long-awaited guidance on the use of copyrighted content for training artificial intelligence (AI). This report, published on May 9, is the third in the Office’s series of reports on AI. Although the report is a “pre-publication” version, it raised tech eyebrows everywhere.
In the report, the Office made clear that when training AI, companies and programmers would need to be careful to avoid infringement. Technology leaders had been hoping to claim fair use for using copyrighted materials to train AI; this would permit them to use copyrighted materials - books, articles, movies, videos, pictures, etc. - to train their models without compensating the holders of the copyright.
The report states that training AI on copyrighted works “clearly implicate[s] the right of reproduction,” meaning that using copyrighted works for such a purpose would presumptively be infringement unless the developer could prove that fair use or a similar defense applied. The report does not completely rule out the possibility that training an AI could be fair use, but it makes clear that a fair use defense will depend on the facts of an individual case, and specific elements - particularly the likelihood that the AI will cause the copyright holder to lose sales, dilute the market, or lose licensing opportunities - will make a fair use defense less likely to succeed.
The report was a relief to copyright holders, and a disappointment to tech companies. It remains to be seen how Perlmutter’s subsequent firing will shape the Office’s attitude toward AI moving forward.
