Human Creativity Reigns: Guidance for Content Creators on What is (and is Not) Protected by Copyright when Using AI

Content creators and their lawyers now have the U.S. Copyright Office’s third of three reports guiding what copyright protects when creators use Artificial Intelligence (“AI”). The report affirms what courts have long held: copyright protection begins and ends with human creativity – machines themselves are not creative authors. Additionally, early court decisions are offering insight into how courts will apply the fair use doctrine to works created using AI, when those authors are accused of copyright infringement. Together, these developments signal both new opportunities and growing risks for creators, developers, and AI companies as copyright law catches up with technological innovations.

The Copyright Office’s guidance on AI is rooted in a longstanding legal doctrine. For over a century, courts have emphasized that copyright law protects only “the production of [humans’] own genius or intellect.”¹ Even when confronted by new and emerging technology, such as photographs, films, video games, music, and software, the common thread of human creative control has remained.² No matter the medium, human authorship is still required for copyright protection. The Copyright Office’s guidance is merely the next iteration of applying well-established principles to the latest new technological developments.

To receive copyright protection for work created using generative AI, protectable elements must reflect human creativity. The use of AI to assist the author in the creative process does not negate protection. Rather, sufficient human contributions to the AI output, such as brainstorming, editing, curating, or changing the AI output with expressive choices, will be copyrightable in whole or in part. Works that are entirely AI-generated, however, fail to be eligible for copyright protection, no matter how detailed the human prompt is. Prompts alone are unlikely to have expressive elements determined by humans. Even if used as a tool, applicants should nonetheless be prepared to disclose the use of AI and explain which elements are human-contributed and how their creative input has shaped the work.

The Copyright Office also signaled that infringement by defendants who use copyrighted works to train AI models can constitute fair use in some cases. At the same time, the Office emphasized that not all uses will be fair use as the analysis depends on specific facts, such as source and use of the training materials and whether the outputs compete with the original works, suggesting licensing as an alternative.

Just weeks after the Office’s pre-publication of the final guidance, the Northern District of California issued one of the first substantive rulings on how fair use applies to generative AI in Bartz v. Anthropic PBC.  In that case, Anthropic compiled millions of copyrighted books, some that it knowingly acquired from pirated sites, into a “central library” to train its large language models (“LLMs”).³ Among the books and works in the central library were those written and/or owned by the author-plaintiffs.⁴ The court granted summary judgment in part to Anthropic, ruling that training LLMs with copyrighted books constituted fair use, finding that the use was “exceedingly transformative.”⁵ akin to the way humans read and re-read books to learn.⁶  The court also found that Anthropic’s act of digitizing printed books to create a central library constituted fair use because the printed copies were discarded, making the new digital copies replacements of the printed copies and not redistribution materials.⁷ However, the court denied Anthropic’s motion for summary judgment on the plaintiffs’ piracy claim, finding that Anthropic’s act of knowingly downloading and copying pirated works was not fair use.”⁸ The case will proceed to trial on this last issue and to determine any resulting damages, and then likely be appealed.

The Bartz decision reaffirms the Copyright Office’s guidance: not all uses of AI are fair uses, and the legality of AI training hinges on how and where the data was used and obtained.

This blog was authored by Stephen Zralek, Anna Schuler, and Yatzaret Aguirre-Sanchez at Spencer Fane LLP.  Stephen, a partner in the Nashville office, and Anna, an attorney in the Overland Park office, help lead the firm’s Creative Content & Brand Protection Market Team, providing a more specialized arena for guidance and collaboration among colleagues navigating creative copyright and trademark disputes and evolving IP regulations as artificial intelligence and other progressive technologies continually reshape the legal landscape. Yatzaret is a summer associate in the Nashville office and a rising 3L at Belmont University College of Law. For more information, visit www.spencerfane.com.

—

¹ _{Burrow-Giles Lithographic Co. v. Sarony, 111 U.S. 53, 58 (1884).}

² _{See, e.g., Sarony, 111 U.S. at 58 (applying copyright law to then-new photographs and cameras); Meshwerks, Inc. v. Toyota Motor Sales U.S.A., 528 F.3d 1258, 1264–65 (10th Cir. 2008) (digital modeling); Stern Elecs., Inc. v. Kaufman, 669 F.2d 852, 856–67 (2d Cir. 1982) (audiovisual work); M. Kramer Mfg. Co. v. Andrews, 783 F.2d 421, 436 (4th Cir. 1986) (video games); Tandy Corp. v. Personal Micro Computs., Inc., 524 F. Supp. 171, 173 (N.D. Cal. 1981) (computer program and silicon chip).}

³ _{Ord. on Fair Use, p. 1, Bartz v. Anthropic PBC, No. 3:24-cv-05417-WHA (N.D. Cal. June 2023, 2025).}

⁴ _Id.

⁵ _{Id. at 9.}

⁶ _{Id. at 12.}

⁷ _{Id. at 9.}

⁸ _Id.