Game of Thrones Meets the U.S. Copyright Office

In a series of plot twists that would make George R.R. Martin proud, the typically staid U.S. Copyright Office has become a proxy in the increasingly heated battle between tech companies developing generative AI products and the creators suing them for copyright infringement, many of whom I'm proud to be representing in several ongoing matters. 

To set the scene: After generative AI burst into public consciousness a couple of years ago, the U.S. Copyright Office began seeking public comment in August 2023 on the myriad copyright issues raised by the fact that Meta and OpenAI, among others, are training their LLMs on millions of copyrighted works. During the lengthy comment period, companies like Meta submitted lengthy, heavily lawyered submissions arguing that they were not committing mass copyright infringement, or, even if they were, it should be considered fair use. On the other side of the ledger, authors and trade groups representing creatives submitted their own views.  

While the Copyright Office waded through the submissions and conducted its own expert analysis, litigation against AI companies began mounting. In recent months, the revelation that Meta sourced much of the copyrighted material it used from illegal pirated databases has added to the debate. Following President Trump's election and Elon Musk's influential role in the administration, observers have speculated that Mark Zuckerberg, who has openly sought to curry favor with the President, would try to get the administration to endorse a stance that training LLMs on copyrighted materials is fair use.

Flash forward to May 9, and the Copyright Office issued its much-anticipated report on Generative AI training, weighing in on many contested issues, often siding with the plaintiffs in the ongoing cases. For instance, the Copyright Office wrote that the use of pirated materials should "weigh against fair use" when analyzing factor one. It further wrote that, under factor four, courts could consider the potential risk of market dilution harm given that LLMs can produce, at scale, works that may compete with those of human authors.

The surprise release of this “pre-publication" report immediately captivated the copyright community. But what came next was even more shocking. Just a day later, the Trump Administration fired Shira Perlmutter, the Register of the U.S. Copyright Office, without explanation. That quickly led to speculation that her firing was retaliation for the office's report. Compounding the confusion, some of President Trump's closest advisers, including Mike Davis, publicly condemned the firing. Davis posted online, "Now tech bros are going to attempt to steal creators’ copyrights for AI profits," which President Trump himself re-posted, amplifying to his millions of followers.

Just yesterday, the intrigue deepened, with Register Perlmutter suing the Trump administration, arguing that her removal was unlawful. 

So what's the takeaway from all these rapid-fire developments?  The fight over AI companies training their LLMs on copyrighted works will continue to play out in both courtrooms and the Executive and Legislative branches. Congress passed the  Copyright Act that courts are now interpreting, and could amend it to clarify some of the open issues with which judges are grappling. A full legislative fix may be unlikely, but in the meantime, politicians and regulators will likely weigh in on both sides. Unlike many hot-button issues, this one doesn't have a clear partisan valence, which may lead to unexpected coalitions. Ultimately, I remain hopeful that courts, legislators, and public opinion will stand with the authors, artists, and other creators whose work is essential both to advancement and progress more generally and to the operation of LLMs specifically.