Does an Open Licence Cover Training an AI? The Question Creative Commons Never Answered

Here is a question the people who built the open web did not have to ask, and now cannot avoid: when a machine consumes millions of Creative Commons works to learn how to write and draw, does the licence on those works permit it? For twenty years the answer to "can I use this?" was reassuringly clear — read the terms, follow them, build something new. The rise of systems that ingest openly licensed material by the million to train generative models has made that once-simple question genuinely hard. This essay is not a verdict; no one has a clean one. It is an attempt to lay out honestly why the question is so difficult, and what it means for a movement built on sharing.

What the bargain actually was

To understand why AI training is unsettling, start with what a Creative Commons licence was designed to do. It is a way for a creator to grant permissions ahead of time — to say "use this under these conditions, without asking me first." The conditions are the familiar set: give attribution, keep the same licence on derivatives, avoid commercial use, do not make derivatives at all. Every one of them presumes a certain kind of user: a person who encounters a specific work, reads the terms, and deliberately complies. Credit this photographer. Keep this remix under the same licence. Do not sell this song.

That presumption is the whole issue. The licences were written to govern reuse as it was understood — copying, adapting, redistributing works you could point to. They were not drafted with any thought of a system that swallows enormous quantities of material, extracts statistical patterns, and emits new work containing no identifiable copy of any single source. This is not an oversight anyone should be blamed for; the technology simply did not exist. But it means that when machine learning meets a Creative Commons licence, it meets an instrument built for a world that no longer fully describes how the work is being used.

Where the conditions stop making sense

Push on the specific licence conditions and the difficulty becomes concrete. Take attribution — by far the most popular Creative Commons requirement. The licence may demand that anyone using the work credit its author. But a model trained on millions of works reproduces none of them individually and cannot, in any meaningful way, attribute the countless sources that shaped a given output. So is the condition broken, met, or simply inapplicable? There is no obvious answer, because the situation the condition was written for — a person reusing an identifiable work — has dissolved.

The same puzzle recurs across the other terms. A creator who chose a non-commercial licence to keep their work out of profit-making hands may feel that using it to train a commercial AI plainly violates that choice; another reader may argue the term governs redistributing the work, not learning statistical patterns from it. Share-alike raises its own version: if a model learns from share-alike works, what exactly would "keep the same licence" even attach to? These are not lawyerly quibbles. They go to the heart of what creators believed they were agreeing to, and reasonable people in the open-culture world land on genuinely different answers.

The part the licence can't capture

Beneath the legal ambiguity is something the text of no licence can settle, and it is where most of the feeling comes from. A great many people released work under Creative Commons out of generosity — a belief in a shared cultural commons, a wish to let others learn from and build on what they made. That generosity was offered to a world of human reusers. Learning that the same openness may also feed commercial systems that could, in aggregate, compete with the very creators who contributed strikes many as a betrayal of the bargain, even where no specific term is clearly broken. The letter of the licence and the spirit in which people accepted it have come apart, and the spirit is the part that made the commons thrive.

This matters far beyond any individual dispute, because Creative Commons worked precisely because creators trusted it — trusted that sharing openly would be honoured in the spirit intended. If a large group of contributors comes to feel their generosity was quietly repurposed for something they never imagined, that trust thins, and with it the willingness to share that made the commons rich. The whole enterprise runs on a virtuous circle of giving, and anything that makes people warier of giving threatens it. It is a cousin of the trust-and-behaviour dynamics we traced in how shared digital culture changed discovery online: change the terms on which people share, and you change whether they share at all.

What openness has to decide now

So the open-culture world is left holding a hard problem, and its response is instructive even though nothing is resolved. Part of the thinking is about clarity going forward — making it explicit whether a work's terms extend to AI training, so creators can express a preference rather than leaving it to interpretation. The instinct is to hand creators back the informed choice the technology took from them, through clearer signals, preference tools, or plain guidance about what existing licences do and do not reach. The aim is less to forbid training outright than to make consent mean something again.

The deeper question underneath is genuinely unresolved: how do you preserve the openness that makes a commons valuable while respecting creators who never wanted their generosity feeding systems they could not foresee? Lean too far toward restriction and you strangle the free reuse that was the point; lean too far toward permission and you drive creators away from sharing at all. Openness and consent, usually allies, are pulling against each other more sharply than ever. There is no tidy answer yet, and anyone offering one is selling a certainty the field does not have. What is clear is that the comfortable assumption of the last two decades — that an open licence meant a settled, human kind of reuse — is over. For anyone weighing how to share their own work amid this, our guide to releasing your own work under a Creative Commons license is still a solid starting point, even as the ground beneath it shifts.

In the end this is a story about a good idea meeting a world its authors could not see coming. The licences were a triumph of their moment; the task now is to carry their spirit — informed, creator-led sharing — into a moment that looks nothing like it. That work is unfinished and contested, and how it goes will decide whether the digital commons stays a place creators actually want to contribute to.