Artists Are Lawyering Up Against AI Slop, and Some Are Even Winning

Kirk Wallace Johnson spent five to six years writing two books — The Feather Thief and The Fishermen and the Dragon — only to discover, when The Atlantic published a searchable dataset of AI training material, that both had been pirated and used to train a chatbot without his knowledge, permission, or compensation. He is not alone: dozens of authors, musicians, and illustrators have taken their complaints to court, and according to multiple reports, some are starting to win.

The Claim

The central accusation across these cases is straightforward: AI companies scraped copyrighted works — books, illustrations, musical compositions — and used them as raw fuel for training large language models and image generators, without asking permission from or compensating the people who created those works. The companies named span the major players: Anthropic, Google, Meta, Stability AI, Midjourney, and others.

Johnson's reaction was pointed. After finding his name in The Atlantic's dataset, he proactively contacted Susman Godfrey, the law firm already leading an active case against Anthropic on behalf of authors. He described the suit as "a middle finger on behalf of everyone that has tried to create something" — a phrase that captures both the personal and economic dimensions of the anger driving these filings.

The legal strategy is not monolithic. Most suits pursue copyright infringement as the primary theory, but some plaintiffs have also pursued terms of service violations as a secondary avenue — an acknowledgment that the copyright question alone may not resolve everything cleanly. Sarah Andersen, the illustrator behind the webcomic Sarah's Scribbles, was among the earliest to file. Along with artists Karla Ortiz and Kelly McKernan, Andersen filed a class action against Stability AI, Midjourney, DeviantArt, and Runway AI starting in January 2023. That case has been crawling through the courts for more than three years.

What We See

Multiple reports confirm this is no longer just a wave of aspirational filings. Some cases have settled, and the fair use defense — the argument that training on copyrighted works constitutes a legal "transformation" of the material — is showing cracks under courtroom pressure. That matters. Fair use has long been the AI industry's most confident first line of defense, and seeing it contested successfully at all represents a genuine shift in the legal landscape.

The Guardian, covering a parallel trend in June 2026, documented artists responding to AI-generated content not just through courts but aesthetically — deliberately embracing imperfection, handcraft, and visible human effort as a counter-signal to what critics now call "slop." The Coca-Cola holiday ad that deployed AI to generate hyperrealistic polar bears and delivery trucks became a widely cited case study in this backlash: widely mocked, it spawned its own news cycle and was remembered more for the public distaste it generated than for the product it promoted. The Guardian's reporting — a separate source from The Verge's legal coverage — adds important context: the legal battle and the cultural rebellion are feeding each other, and participation is high because creators see their economic futures at stake, not just their past work.

Our read is that the litigation pressure is doing something public discourse alone could not: forcing specificity. When courts examine whether a particular model trained on a particular dataset infringed a particular author's particular books, the abstract "AI training is transformative" argument must be defended with actual evidence in a specific factual record. That is a much harder fight than issuing a policy statement, and it appears to be one the industry is losing in at least some instances. The beSpacific summary of the Verge reporting, published July 28, 2026, frames the mix of anger, financial worry, and desire for accountability that Johnson described as broadly representative — not one author's idiosyncratic grievance, but a sentiment shared across media types and career stages.

Where It Falls Short

The picture is murkier than the headline "some are winning" implies. The SaveDelete summary of the same Verge piece notes that outcomes have been mixed, with wins on specific issues interspersed with losses — particularly around how different courts interpret fair use in different factual contexts. A settlement is not the same as a judicial finding that training on copyrighted data constitutes infringement; companies settle for strategic reasons that have nothing to do with legal weakness. Until an appellate court or the Supreme Court produces a definitive ruling, every lower-court outcome remains provisional and jurisdiction-specific.

There is also a scope gap worth naming explicitly. The cases filed so far target training data — the inputs used to build these models. A separate and still largely unlitigated question concerns output liability: whether AI systems that generate content substantially similar to a specific copyrighted work expose their operators to infringement claims on the output side. That question sits one layer removed from where current litigation is concentrated, and its resolution will matter enormously for developers and companies building products on top of these APIs. Training-data victories for artists, even significant ones, may not settle what happens when a deployed model generates something that looks more like copying than inspiration.

The Anthropic case, led by Susman Godfrey, is among the most closely watched precisely because Anthropic has positioned itself as the industry's safety-focused, responsible-development player. A finding against Anthropic on training data would carry symbolic weight well beyond any dollar amount — it would signal that no company in this space, regardless of its stated values, has insulated itself from liability by behaving carefully after the training data was assembled.

Sources