fair use

Fair use is a legal exception that allows someone to use another person's book, art, or music for certain purposes without asking permission first. Think of it like borrowing a book from a library, reading it, and quoting parts of it to write a review. You're not copying the whole book to sell it — you're taking only what you need to create something entirely different, which is why it isn't treated as copyright infringement.

Courts generally weigh three things when deciding this: what the material was used for, how much of it was taken, and whether the result competes with the market for the original work. Recently, training AI on books and papers has been put to this test. Courts have found that training itself counts as fair use, since the AI isn't reproducing and reselling the original work but reading vast amounts of text to produce something different. However, how the material was obtained is treated as a separate issue — if pirated copies were downloaded without authorization, that part can still carry liability.

News coverage often mentions "the ruling ordering Anthropic to pay $1.5 billion," which can be misread as meaning AI training was found to infringe copyright. In fact, what the judge ruled unlawful wasn't the training itself but how the books were obtained — downloaded without authorization from pirate sites. Using copyrighted works as training material was itself recognized as fair use. The same ruling contains two conclusions at once: training is legal, but the way the material was acquired was not.