How Is the U.S. Copyright Office Dealing With AI?

The U.S. Copyright Office has recently released a significant report addressing the complex relationship between artificial intelligence (AI) and copyright law, particularly focusing on the use of copyrighted materials in training generative AI models.That report is pretty long, over 100 pages, but inside it are some key findings that I think should be really eye-opening to creatives from every walk of life. Let's dive in. Key Takeaways From the AI Copyright ReportRight before that big report was filed, the Trump administration fired Dr. Carla Hayden, the Librarian of Congress. Then, a day after the report came out, the Register of Copyrights, Shira Perlmutter, was fired.Now, I'm not saying this report led to all of that, but I am saying that no matter who is in power, we need to make sure they have the backs of the creatives and not of the big tech companies in these AI situations. I read Copyright Lately's analysis of the report and compiled some takeaways that I think our audience needs to know about.Copying Copyrights Starts EarlyThe act of AI copying copyrighted material happens during the initial stages of AI model training. And by the time it's trained, the elements of these copyrighted works may stay within the AI model, which can potentially lead to copyright infringement in the future for other projects. Basically, if the machine learns from your stuff, it remembers it and can use it over and over again.Transformative Use Depends on ResultsThe whole idea is that these AI models are learning from our copyrighted stuff, but we can't do anything about it because by the time it regurgitates it, it's different.The report suggests that determining whether the use of copyrighted material in AI training qualifies as "transformative" under fair use principles will depend heavily on how the resulting AI model is ultimately used. Like, if you have an animation character and then it creates one similar, sort of like the Studio Ghibli experiment, then you probably can sue. But if it's trained on the Ghibli characters but spits out something different, then you probably can't do anything about it. It really has to do with what gets spit out. Some applications of this might be seen as highly transformative, while others could be viewed as merely substituting the original copyrighted work.Rejection of "Non-Expressive" Use One of the most interesting points in the report was about how humans learn. Basically, we have to open all the books and see what we retain to learn. But AI just goes in there, copying all the books and taking all the info from them. The Copyright Office has pushed back against arguments that AI training constitutes a "non-expressive" use or that AI simply mimics human learning processes. We know that it doesn't. But we have to keep an eye on what the current White House believes and thinks, since they're making changes all the time. The report highlights fundamental differences in how AI systems process and retain information compared to humans.It says on page 48, “Humans retain only imperfect impressions of the works they have experienced . . . Generative AI training involves the creation of perfect copies with the ability to analyze works nearly instantaneously.” Copying Entire Works This has been what I have been most worried about as a writer. I don't want AI with the ability to copy everything I've done. AI companies argue fair use, in that if it's on the internet, they have the freedom to use it to learn. So, let's get into the nuance. For AI, copying the entirety of a work is functionally necessary for it to be able to produce anything. It cannot create something "transformative" without copying everything. So you have a roadblock, if it can't get to the non-infringement end without infringing, should it be allowed to infringe? That's one of the big issues facing all of AI right now. Summing It All Up We're in an interesting time as creatives. AI is this fun thing that's supposed to be helpful to us, making our dreams come true and streamlining the process, along with all the other benefits they're selling. But it's also ripping us off left and right, and they have a lot of money behind it, bending laws and pressuring politicians to make sure they can get away with it. All we can do is stay informed and try to fight this stuff where we can. Let me know what you think in the comments.

May 13, 2025 - 22:02

How Is the U.S. Copyright Office Dealing With AI?

The U.S. Copyright Office has recently released a significant report addressing the complex relationship between artificial intelligence (AI) and copyright law, particularly focusing on the use of copyrighted materials in training generative AI models.

That report is pretty long, over 100 pages, but inside it are some key findings that I think should be really eye-opening to creatives from every walk of life.

Let's dive in.

Key Takeaways From the AI Copyright Report

Right before that big report was filed, the Trump administration fired Dr. Carla Hayden, the Librarian of Congress. Then, a day after the report came out, the Register of Copyrights, Shira Perlmutter, was fired.

Now, I'm not saying this report led to all of that, but I am saying that no matter who is in power, we need to make sure they have the backs of the creatives and not of the big tech companies in these AI situations.

I read Copyright Lately's analysis of the report and compiled some takeaways that I think our audience needs to know about.

Copying Copyrights Starts Early

The act of AI copying copyrighted material happens during the initial stages of AI model training.

And by the time it's trained, the elements of these copyrighted works may stay within the AI model, which can potentially lead to copyright infringement in the future for other projects. Basically, if the machine learns from your stuff, it remembers it and can use it over and over again.

Transformative Use Depends on Results

The whole idea is that these AI models are learning from our copyrighted stuff, but we can't do anything about it because by the time it regurgitates it, it's different.

The report suggests that determining whether the use of copyrighted material in AI training qualifies as "transformative" under fair use principles will depend heavily on how the resulting AI model is ultimately used. Like, if you have an animation character and then it creates one similar, sort of like the Studio Ghibli experiment, then you probably can sue.

But if it's trained on the Ghibli characters but spits out something different, then you probably can't do anything about it. It really has to do with what gets spit out. Some applications of this might be seen as highly transformative, while others could be viewed as merely substituting the original copyrighted work.

Rejection of "Non-Expressive" Use

One of the most interesting points in the report was about how humans learn. Basically, we have to open all the books and see what we retain to learn. But AI just goes in there, copying all the books and taking all the info from them.

The Copyright Office has pushed back against arguments that AI training constitutes a "non-expressive" use or that AI simply mimics human learning processes.

We know that it doesn't. But we have to keep an eye on what the current White House believes and thinks, since they're making changes all the time.

The report highlights fundamental differences in how AI systems process and retain information compared to humans.

It says on page 48, “Humans retain only imperfect impressions of the works they have experienced . . . Generative AI training involves the creation of perfect copies with the ability to analyze works nearly instantaneously.”

Copying Entire Works

This has been what I have been most worried about as a writer. I don't want AI with the ability to copy everything I've done.

AI companies argue fair use, in that if it's on the internet, they have the freedom to use it to learn.

So, let's get into the nuance.

For AI, copying the entirety of a work is functionally necessary for it to be able to produce anything. It cannot create something "transformative" without copying everything. So you have a roadblock, if it can't get to the non-infringement end without infringing, should it be allowed to infringe?

That's one of the big issues facing all of AI right now.

Summing It All Up

We're in an interesting time as creatives. AI is this fun thing that's supposed to be helpful to us, making our dreams come true and streamlining the process, along with all the other benefits they're selling.

But it's also ripping us off left and right, and they have a lot of money behind it, bending laws and pressuring politicians to make sure they can get away with it.

All we can do is stay informed and try to fight this stuff where we can.

Let me know what you think in the comments.