The defining intellectual-property fight of the decade is moving through courtrooms at the speed of law, which is to say slowly, expensively and with consequences that will long outlast the news cycle. Authors, illustrators, photographers, musicians and major news organisations have sued leading AI companies, alleging that models were trained on copyrighted work without permission and can reproduce its substance on demand. The defendants answer that training is fair use, transformative and no more infringing than a human reading widely and writing originally.
Everything turns on analogies, because the technology fits no existing box. Is training a model like a student reading a library, or like a photocopier with a memory? Is the model itself a derivative work, or a tool like a camera, innocent until aimed? Courts in different jurisdictions are reaching for different analogies, and the resulting map will decide business models worth hundreds of billions. We follow the docket in our law coverage, where the future of creative industries is being footnoted.
Licensing is racing the litigation
While judges deliberate, the market is hedging. Major publishers and image libraries have signed licensing deals with AI firms, creating a paid lane alongside the litigated one, and the deals themselves become evidence: of a market for training rights, say the plaintiffs; of good-faith transition, say the defendants. The US Copyright Office has added its own analysis, concluding among other things that purely machine-generated material, without human authorship, cannot be copyrighted, a ruling that shapes the output side of the argument even as the input side is fought over.
The settlement horizon
Legal veterans caution that many landmark technology cases end not with a trumpet blast of precedent but with settlements, licensing frameworks and exhausted litigants agreeing on a price for peace. The music industry’s wars over file-sharing followed exactly that arc, from existential lawsuits to the licensed streaming economy. A similar landing zone is visible here: per-work or per-catalogue training licences, opt-out registries with teeth, and revenue-sharing experiments for generated content in a recognisable style.
What the courts will still decide, settlement or not, is the default. If training on unlicensed work is lawful, licensing is a courtesy; if it is not, licensing is the market. On that single default rests the bargaining power of every creator alive, which is why the footnotes matter more than the headlines.
Why creators should care about the boring parts
The headline cases involve famous names, but the precedents will govern everyone: the illustrator whose style is mimicked, the local paper whose archive trained a chatbot, the musician whose catalogue is suddenly a dataset. Copyright was built, over centuries, to balance the incentive to create against the public’s interest in building on creation. AI has not repealed that balance. It has forced the law to find it again, in public, under time pressure, with the whole creative economy watching the footnotes.
