Copyright Office Offers New Guidance on AI

In June, the US Copyright Office hosted an online event focusing on the registration of works containing generative artificial intelligence (GAI) content.

We previously wrote in this blog about how a comic book creator sought and initially received copyright registration for an 18-page comic book titled Zarya of the Dawn. After she revealed on Instagram that the artwork had been created by a GAI tool called Midjourney, and a reporter contacted the Copyright Office, the Office concluded that “the images in the Work that were generated by the Midjourney technology are not the product of human authorship” and thus not eligible for copyright protection.

As we discussed here, the Copyright Office subsequently issued new copyright registration guidance. This makes clear that copyright registration applicants must disclose the inclusion of GAI content in works submitted for registration.

At the online event, the Copyright Office staff discussed the new guidance and how it will be applied.

One slide at the presentation noted that “Disclosures have always been required if the work contains unclaimable material.” For example, a registered work might include work that was previously published or registered, work in the public domain, and work owned by third parties and used with permission or under “fair use” principles.

Copyright registration only applies to the original portion of such combined works.

Failing to disclose that an AI tool created all or part of a work submitted for copyright registration can subject the applicant to fines under 17 U.S.C. §506(e) for making “a false representation of a material fact.”

However, the Copyright Office isn’t checking prior registrations to see whether they include AI material (assuming they could even tell).

The Office noted that, in order to be registered, a work must pass the de minimis test for human authorship set forth by the US Supreme Court in the case of Feist v. Rural Telephone.

Feist was a 1991 case involving paper telephone directories. When one directory company refused to license its white pages listings to Feist, Feist extracted the information it needed from the other directory without permission and published its own book.

The Supreme Court held that white pages (which contain only names, addresses, and phone numbers) are not entitled to copyright protection, and thus copying them is not copyright infringement.

The Court noted that

(a) Article I, § 8, cl. 8, of the Constitution mandates originality as a prerequisite for copyright protection. The constitutional requirement necessitates independent creation plus a modicum of creativity.

Facts aren’t original or creative, and thus can’t be copyrighted.

Section 313.4(B) of the Compendium of U.S. Copyright Office Practices further states that

Works that contain no expression or only a de minimis amount of original expression are not copyrightable and cannot be registered with the U.S. Copyright Office.

According to the new Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence, released in March,

AI-generated content that is more than de minimis should be explicitly excluded from the copyright application.

(The issue then becomes, how much is “de minimis”?)

The Office suggested that it would allow the registration of a song by the Beatles that used AI-powered mastering technology.

As The Guardian reported, Sir Paul McCartney has announced a new and final Beatles song will be released later this year. This will use AI to “extricate” John Lennon’s voice from an old demo.

According to The Guardian,

Though McCartney did not name the song, it is likely to be a 1978 Lennon composition called Now and Then. The demo was one of several songs on cassettes labelled “For Paul” that Lennon made shortly before his death in 1980, which were later given to McCartney by Lennon’s widow, Yoko Ono.

On the other hand, the Office suggested that the AI elements of a television opening credits sequence would need to be disclosed and excluded from copyright registration.

As TechCrunch reported, artists were upset that the opening credits for Secret Invasion, Marvel’s latest streaming series on Disney+, were designed by VFX company Method Studios using artificial intelligence.

According to Marvel,

No artists’ jobs were replaced by incorporating these new tools; instead, they complemented and assisted our creative teams.

TechCrunch noted that

Artists are increasingly worried that AI creations will not only discredit their work but also threaten their livelihoods. Generative AI takes millions of images made by actual people to create a hodgepodge of art that once took someone hours, days, months or even years to produce.

Jeff Simpson, who worked as a visual development concept artist on Secret Invasion, said on Twitter that “I believe AI to be unethical, dangerous and designed solely to eliminate artists’ careers.”

GAI is based on “prompts” usually provided by humans. For example, a prompt might be “write me a 1000-word blog about copyright and AI” or “write me a song about pickles in the style of the Beatles.”

At the online event, Rob Kasunic, Associate Register of Copyrights and Director of Registration Policy and Practice, was asked whether GAI prompts could be copyrightable. He said that it’s technically possible but that he wasn’t aware that anyone had yet tried to register a prompt.

Also, the registration would only apply to the language of the prompt, and not to any output (such as text or images) created by AI based on the prompt.

This area of the law is still very much in flux, but this suggests that an AI prompt might be protected by copyright law much as software code is now.

For example, directly copying the code that enables Microsoft Word would be copyright infringement. However, other companies are free to come up with their own word processing software that accomplishes the same things that Word does.

But there are only so many ways to write a prompt like “write me a song about pickles in the style of the Beatles.” So will it be possible to monopolize the output for such a prompt by registering all possible inputs? Presumably, the Copyright Office will soon have to resolve this question.

A public webinar called “International Copyright Issues and Artificial Intelligence” is scheduled for July 26 and will cover AI authorship, training, and infringement.

More information on announcements, events, and resources related to AI and copyright are at copyright.gov/ai.