The year the lawsuits caught up to the technology

Generative AI spent late 2022 as a party trick. By the start of 2023, it was a defendant.

In January, visual artists Sarah Andersen, Kelly McKernan, and Karla Ortiz filed a class action against Stability AI, Midjourney, and DeviantArt in the Northern District of California, arguing that the companies had trained their image generators on millions of copyrighted works without permission (Copyright Alliance). Weeks later, Getty Images filed its own suit against Stability AI in Delaware federal court, alleging the company had used more than 12 million Getty photographs, "image quality, variety of subject matter and detailed metadata" included, to train Stable Diffusion without a license (Brittain).

Neither case was about whether AI image generators work. Both were about how the training data got there in the first place, and whether "we scraped the open internet" is a defense or an admission.

The courts have not been fully persuaded by either side yet. On October 30, a judge dismissed most of the artists’ claims against Midjourney and DeviantArt while allowing the core copyright infringement claim against Stability AI to move forward, with leave to amend (Copyright Alliance). That is not a win lap for anyone. It is a signal that the legal system is still building the vocabulary to answer a question this new. But the cases themselves already did something the technology hadn’t: they put "consent" back in the center of the conversation, right where it belongs.

What the artists are actually asking for

It would be easy to flatten this into a fight about whether AI is good or bad. It isn’t that. Listen to the people who filed these suits and a narrower, more specific demand comes through: not a ban, a boundary.

Karla Ortiz said it plainly in testimony before the Senate Judiciary Subcommittee on Intellectual Property this July, when lawmakers convened a hearing on AI and copyright law.

“These AI companies use our work as training data and raw materials for their AI models without consent, credit, or compensation.”

Karla Ortiz, written testimony before the U.S. Senate Judiciary Subcommittee on Intellectual Property, July 12, 2023

Three words carry that whole sentence: consent, credit, compensation. Not one of them requires generative AI to stop existing. All three require the companies building it to ask first instead of asking forgiveness later, and to pay the people whose work made the model possible in the first place. That is not an anti-technology position. It is a labor position, and it deserves to be treated as one.

Hollywood fought the same fight, in public, and won ground

While the lawsuits worked their way through federal court, the same argument played out on a bigger stage. The Writers Guild of America went on strike May 2 and stayed out for 148 days, one of the longest work stoppages in the union’s history, with AI protections as one of its central demands (Kilkenny and Goldberg). The contract members ratified on October 9, with 99 percent approval, secured real terms: studios cannot use AI to write or rewrite literary material, cannot use a writer’s work to train AI without consent, and cannot use AI-generated material to undercut a writer’s credit or pay (Kilkenny and Goldberg).

SAG-AFTRA held the line right behind them. The union struck July 14, and as recently as November 6, two days before this piece was written, its own negotiators said AI remained one of the last unresolved issues at the table. Then, in a unanimous vote, the union’s negotiating committee approved a tentative agreement, ending a 118-day strike at 12:01 a.m. this Thursday (Cerullo).

Two of the largest unions in Hollywood spent the better part of a year treating AI consent as something worth walking off the job over. That is not overreaction. That is what conviction looks like when the stakes are someone’s livelihood instead of someone’s headline.

Consent is not a technicality, it is the whole question

Here is what makes 2023 different from every earlier wave of "the robots are coming for creative jobs" anxiety: this time, the industry has proof that speed and consent don’t have to be opposites.

McKinsey’s research on generative AI adoption, published this August, found that a third of surveyed organizations were already using generative AI regularly in at least one business function, less than a year after the tools most people know went public (McKinsey & Company). That is a genuinely fast adoption curve. It is also exactly why the question of what trained these tools, and who agreed to it, cannot be an afterthought bolted on once the lawsuits arrive.

It can be done differently. When Adobe introduced Firefly this spring, it built the first model on Adobe Stock images, openly licensed content, and public domain material where copyright had already expired, and it committed publicly to building a compensation model for the Stock contributors whose work trained it (Adobe). We are not naive about the difference between a press release and a finished payment system, and Adobe’s own contributors have pushed back hard on how that promise has played out in practice. But the architecture Adobe chose, licensed and disclosed instead of scraped and defended, proves the industry had a consent-first option available the entire time. Most of it didn’t take it. That was a choice, not an inevitability.

Where HAUS draws the line

We’ve said this before and we’ll keep saying it until it stops needing to be said: AI sits at our table. It doesn’t run the haus.

That belief only means something if it survives contact with a hard year like this one. So here is the specific version. We don’t run creative work through tools trained on data nobody agreed to hand over. We don’t treat "the model can do it" as a reason to stop paying a human for the judgment that makes it good. And when we use AI in our own process, we say so, out loud, because the issue was never the tool. It’s misrepresenting how the work got made.

“A tool that can’t say where it learned everything it knows isn’t a shortcut. It’s a liability wearing a shortcut’s clothes.”

David Keyes, Founder & CEO, HAUS XXIV

That standard costs us something sometimes. A consent-first tool is occasionally slower, or narrower, or less flashy than the version built on everything the open internet ever published without asking. We’ll take slower. Safe doesn’t survive here, but neither does theft dressed up as innovation. The artists who filed suit this year and the writers and actors who walked off set weren’t asking the industry to slow down out of fear. They were asking it to grow up. We’d rather build alongside creatives who insisted on that than build fast on top of the ones who didn’t get a say.