There is a passage in Drawn to Extinction where I describe Disney striding, not creeping, into a licensing deal with OpenAI, handing over more than two hundred characters from Disney, Marvel, Pixar and Star Wars so that anyone with a subscription and a spare afternoon could prompt them into short videos, with a billion dollars of equity on top. I wrote that story had become asset and asset had become dataset, and I still remember the realisation of witnessing the thing you have warned about for two years arrive as a fancy press release and a smug Sam Altman social media post.
Then, in March this year, OpenAI switched Sora off. The app, the tools and the developer interface wound down inside a few weeks, with the company saying it wanted to redirect itself towards business processes instead, and the Disney partnership went with it before any money changed hands. Disney said, with the sort of glacial courtesy that tells you exactly how furious everyone was behind the door, that it respected OpenAI’s decision to exit the video generation business.
My first reaction was relief and there is a version of me that wanted to punch the air about the whole thing collapsing under its own compute bill. My second, which has not left since, was that the collapse changes almost nothing about the position working comics creators are in. The deal was never the mechanism. Enforcement is the mechanism, and enforcement is where the real story of the past eighteen months has been hiding while everybody watched the headlines about training data.
Consider what happened in the London courts last November, in the first proper British judgment on generative Ai and copyright. Getty Images had gone after Stability AI with a claim that looked like it might settle the whole argument. By the end of the trial Getty had abandoned its primary copyright claim and its database rights claim, because it could not establish that the training had happened in the United Kingdom, which is a jurisdictional technicality with the force of a demolition charge. What remained was rejected on the basis that the model does not store reproductions of the works it learned from, and so is not an infringing copy of anything. What Getty won, after years and a fortune in costs, was a narrow finding on trade marks, because older versions of Stable Diffusion had been producing images with mangled Getty and iStock watermarks in them.
Sit with that, because it is the whole argument in miniature. One of the largest image licensing companies on earth took the fight to a model developer and the only thing the law reliably protected was the branding, rather than the photograph, the composition, the labour, or the decades of curation sitting behind any of it.
That is not bad lawyering, it is the shape of the protection that exists. Trade marks, characters, likenesses and brands sit in a tier that is genuinely enforceable, because it is held by organisations with legal departments and the appetite to spend three years arguing about disclosure schedules. Individual expression and personal style sit in a tier that is close to unenforceable, because the people holding it cannot fund a federal case, frequently never registered the work, and across enormous swathes of comics history never owned it at all.
You can watch the same fault line open in the American artists’ class action brought by the cartoonist Sarah Andersen and others, still grinding through the Northern District of California. The case has survived, and that matters. But at the first hurdle the court dismissed the claims of class members who had not obtained registrations, because registration is a precondition of bringing the claim at all, which quietly removes an enormous amount of comics labour from the room. Anthology work, self-published minis, webcomics, colouring, lettering, and the whole informal apprenticeship layer where people learn the craft by doing it badly for years before anyone pays them properly.
The British government had its chance in March and changed nothing, dropping the text and data mining exception and the opt-out without adopting anything in their place, so no comics artist has any more route to discovering their work was used than they had a year ago.
So here is the thing that actually keeps me up, and it is not a court case.
Dark Horse has had contractual language since 2024 requiring creators to warrant that their work contains no material generated by computer Ai programs, which is a reasonable position and one I am glad exists. The natural next clause, and I want to be clear that this is my prediction rather than something I can cite, is an indemnity, where the creator agrees not merely to warrant that the work is human but to cover the publisher against any claim arising from it. At that point a freelancer with no legal budget and a page rate that has barely moved in fifteen years is underwriting a publisher’s exposure to a doctrine no court has settled, over tools now baked into half the software on their laptop whether they asked for them or not.
That is the risk. Not the theft, which has happened and cannot be undone, but the compliance burden about to be handed down to the least protected people in the chain while the scrapers settle in bulk and license upwards.
I should be fair to the other side of my own argument. A judge in San Francisco has sketched out a theory called market dilution, which holds that flooding a market with machine-made substitutes is precisely the harm fair use analysis exists to prevent, and run properly it could do far more damage than anything Getty attempted. Norms may also outrun law here, because comics is small enough that reputation still functions as enforcement. San Diego banned Ai artwork from its art show outright this year, and Jim Lee said no in front of a cheering hall and meant it.
But norms protect the visible, and the ladder I have spent a book describing runs from the invisible upwards. Which is why I have stopped hoping a judge will eventually understand what a colourist does, and started thinking about four unglamorous things within reach of a working creator this year.
The first is to read the rights grant in your next contract properly, looking for language covering all media now known or hereafter devised, or anything mentioning machine learning or model development. That phrasing has already appeared in book publishing agreements and will reach comics soon enough, and it is the quietest way imaginable to hand over the thing you are worried about losing. You are allowed to strike it, and to ask what it is for.
The second is to look for the word indemnify in the Ai warranty clause, which carries a completely different weight from the word warrant. Warranting that your work is human is a promise, while indemnifying a publisher means funding their defence if somebody accuses you, which a freelancer on a page rate cannot absorb. Ask for a cap, or for it to be limited to your fee, because authors negotiate these routinely and nobody thinks it rude.
The third is to keep your process, not as an artistic principle but as evidence, which means layered files, dated backups and sketchbook photographs. San Diego has made a single coordinator the sole judge of what counts as human, and when somebody points at your page the burden of disproving it will land on you rather than on them. The artist who kept the working files settles that in an afternoon, and the artist who flattened everything spends a fortnight being torn apart by strangers.
The fourth, if you work in or sell into the United States, is to register what you own, because it costs little, it is tediously easy, and it is the literal gate on the courthouse door. The artists whose claims were thrown out in California were not defeated on the merits of anything, they had simply not filed the paperwork.
None of that is a solution, and all four together will not stop a single model training on a single page. What they will do is decide who has standing and leverage when the settlements land, and those are the only currencies that will be accepted in the room where this gets decided.
Because the arrangement now forming will ask the working artist to prove that they are human, at their own cost, on their own time. It will ask the company that scraped them for a cheque.
Sources
Getty Images (US) Inc & Ors v Stability AI Ltd [2025] EWHC 2863 (Ch), judgment of Mrs Justice Joanna Smith, 4 November 2025: https://www.judiciary.uk/judgments/getty-images-v-stability-ai/
Case analysis of the abandoned claims and the narrow trade mark finding, Carson McDowell: https://carson-mcdowell.com/news-insights/insights/getty-images-v-stability-ai-2025-ewhc-2863-ch
Report on Copyright and Artificial Intelligence, Department for Science, Innovation and Technology and Department for Culture, Media and Sport, March 2026: https://assets.publishing.service.gov.uk/media/69ba692226909a14239612e4/CP2602959_-_Report_on_Copyright_and_Artificial_Intelligence_web.pdf
Analysis of the March 2026 report and the dropped text and data mining exception, Bird & Bird: https://www.twobirds.com/en/insights/2026/uk/copyright-,-a-,-aiin-the-uk-the-debate-rolls-on
House of Lords Communications and Digital Committee report on AI, copyright and the creative industries, 6 March 2026: https://committees.parliament.uk/committee/170/communications-and-digital-committee/news/212361/
OpenAI and Disney licensing agreement, December 2025: https://openai.com/index/disney-sora-agreement/
OpenAI shuts down Sora and Disney exits the partnership, Variety, 24 March 2026: https://variety.com/2026/digital/news/openai-shutting-down-sora-video-disney-1236698277/
Jim Lee on DC and generative Ai at New York Comic Con, Tom’s Guide: https://www.tomsguide.com/ai/dc-comics-just-made-its-position-on-generative-ai-crystal-clear-not-now-not-ever
San Diego Comic-Con bans Ai artwork from the 2026 art show, TechCrunch, 25 January 2026: https://techcrunch.com/2026/01/25/science-fiction-writers-comic-con-say-goodbye-to-ai/
Background on the artist-led pressure behind the Comic-Con reversal, Futurism: https://futurism.com/artificial-intelligence/comic-con-ai-art
Disney Enterprises Inc v Midjourney Inc, 2:25-cv-05275 (C.D. Cal.), docket: https://www.courtlistener.com/docket/70513159/disney-enterprises-inc-v-midjourney-inc/
Overview of the current US Ai copyright cases, including Andersen v Stability AI, Bartz v Anthropic and the market dilution reasoning in Kadrey v Meta, Norton Rose Fulbright: https://www.nortonrosefulbright.com/en/knowledge/publications/ce8eaa5f/ai-in-litigation-series-an-update-on-ai-copyright-cases-in-2026