Bartz v. Anthropic: Did Creators Really Lose?

Disclaimer: As usual....I am not an attorney or intellectual property professional. This article reflects my personal opinions and observations as a photographer, artist, and small business owner whose copyrighted work has been infringed multiple times. It is not legal advice.

If you only read the headlines following Bartz v. Anthropic, you probably came away with one conclusion:

"A federal judge ruled AI training is fair use."

For many creators, that sounded like the end of the conversation and like we have no power going forward.

But not me....I don't think that's the full story.

In fact, I think Bartz may be remembered less as a sweeping victory for AI companies and more as a roadmap for what not to do.

What the Court Actually Said

Judge William Alsup concluded that Anthropic's use of lawfully acquired books to train its language model qualified as fair use under the specific facts presented in the case.

That probably felt like a big win for Anthropic.

But that's where many news articles stopped.

The opinion did not say AI companies can copy anything they want.

It did not declare that every AI training dataset is protected by fair use.

And it certainly did not approve the use of pirated works.

Instead, the court carefully distinguished between different acts that had often been discussed as though they were one.

The Important Distinction

One of the most interesting aspects of the decision was that the court analyzed the acquisition of the books separately from the training process.

Anthropic had downloaded millions of books from unauthorized "shadow libraries."

Judge Alsup held that those pirated copies were not protected by fair use.

In other words, while the court found that training on lawfully acquired books could qualify as fair use, it drew a clear line when it came to obtaining copyrighted works through piracy.

That distinction is easy to miss, but it may ultimately become one of the most important parts of the opinion.

It's what led to a huge settlement for using pirated books to train its AI.

Could Bartz Influence the Google Gemini Lawsuit?

One of the first places we'll likely see the ripple effects of Bartz is in the recently filed lawsuit against Google over its Gemini AI models.

Unlike Bartz, which focused on Anthropic's use of books to train Claude, the new lawsuit alleges that Google willfully copied millions of copyrighted books, including works provided through Google Books, Google Play Books, and other sources, to develop Gemini without permission or compensation.

The plaintiffs also allege Google exceeded the scope of existing agreements by repurposing books for AI training and knowingly exposed itself to potentially enormous copyright liability.

That immediately raises questions familiar to anyone who followed Bartz.

Were the books lawfully obtained?

Were they copied beyond the scope of any license or agreement?

Did Google simply possess lawful copies, or did it create and retain unauthorized copies for AI development?

Those questions may become just as important as whether the training itself qualifies as fair use.

The lawsuit alleges that Google did more than train Gemini on books it was authorized to access for limited purposes. According to the complaint, Google also used books that were subject to narrow licensing restrictions in ways the plaintiffs contend exceeded those licenses, while simultaneously obtaining and using pirated copies of copyrighted books to develop its AI models. If those allegations are proven, the case could test many of the same questions raised in Bartz: whether the manner in which copyrighted works are acquired and the scope of any license are just as important as the training process itself.

If Bartz taught us anything, it's that courts may no longer treat "AI training" as one single legal act. Instead, they may analyze acquisition, copying, storage, licensing, and training separately. If that's the framework future courts adopt, companies may face liability for how they build their datasets even if they ultimately prevail on certain fair use arguments.

Whether Google can distinguish its conduct from Anthropic's remains to be seen. But one thing is becoming increasingly clear: AI copyright litigation is shifting away from broad philosophical debates and toward the specific facts of how copyrighted works were acquired, copied, stored, licensed, and ultimately used.

And in my opinion, if we are doing that, we should be opening the door to discuss how our laws change alongside technology. We did with the DMCA, and we should do it now with AI training. More on that below.

The Settlement That Changed the Conversation

Following the ruling, Anthropic agreed to a settlement reportedly valued at approximately $1.5 billion to resolve claims relating to the pirated books.

While the court clarified companies can use legally acquired books for training, Anthropic had to know where the case was going with the stolen material.

If the case had truly established that AI companies could freely copy copyrighted works without consequence, why would one of the largest copyright settlements in history follow?

It set boundaries. Maybe not a boundary we all hoped for, but at least they have to pay us pennies before they use our work to make billions, amirite?

A Roadmap... or a Warning?

If I were advising an AI company, I'd read Bartz as a warning

The court effectively separated two very different questions:

• How did you obtain the copyrighted works?

• How did you use them?

Those questions may sound similar, but legally they are not the same.

The opinion suggests that courts may evaluate the acquisition, copying, storage, and training stages independently rather than treating "AI training" as a single act.

That could have enormous implications for future cases. Personally, it will for my future legal cases, I will specifically point to this case with my attorneys and say, "Let's rememeber this for discovery....."

What About Photographs?

Photographers should be cautious about drawing broad conclusions about photographs because Bartz involved books.

Photographs present different factual and legal issues.

Image-generation models are trained differently than large language models.

Photographers often rely heavily on licensing their work.

Image models also raise questions about visual similarity, derivative works, and whether generated outputs compete directly with existing photographs or illustrations.

Those issues remain largely unresolved.

The ongoing lawsuits involving Stability AI, Midjourney, Getty Images, Disney, Universal, and groups of visual artists will likely answer questions that Bartz never addressed.

Did Creators Really Lose?

I don't think the answer is as simple as yes or no.

Anthropic won an important ruling on fair use for training under the facts presented.

On books.

The authors succeeded in establishing that pirated copies were not insulated by that ruling.

Perhaps most importantly, the case demonstrated that courts are willing to examine AI systems piece by piece rather than accepting broad arguments that everything AI companies do is either lawful or unlawful.

My Take

I understand why many creators might feel discouraged after reading the headlines, especially if they don't dig a little deeper and read the facts.

But I don't believe Bartz closed the door on creators' rights.

If anything, it highlighted just how fact-specific these cases are. How we are still analyzing and challenging what is right and relying on courts to make it clear.

Personally, I think generative AI is a new technology which often requires new rules. Like the Digital Millenial Copyright Act "DMCA" reshaped digital infringement, a new law regarding the use of works for training would be ideal. Something that prohibits the use of large scale training models without a license. That would be fair. If I were an AI company who respected artists, it would be something I'd bake into the company's core values: we compensate the human minds that help build our brand....

Seems too simple, too fair....why not?

As it stands, future courts will still have to answer difficult questions about photographs, artwork, music, newspapers, video, licensing markets, memorization, and AI-generated outputs.

So rather than seeing Bartz as the final chapter in AI copyright law, I see it as the beginning of a much more detailed conversation, hopefully one that leads to looking at human artists as humans first and then placing a value on their involuntarily contributions to billion dollar AI companies.