“There’s a need for guidance and insight when it comes to the use of AI and ML technologies, including generative AI technologies, as evidenced by the more than 150 audience questions received during the live stream. The Office also noted that 1,500 people attended the virtual meeting.”
On Wednesday, June 28, the United States Copyright Office (USCO) hosted a virtual event exploring guidance for registration of works containing generative artificial intelligence (AI) content. The hour-long event included a recap of the USCO’s previously released policy guidance and the Zarya of the Dawn partial registration refusal, staff walking through numerous examples of how AI technologies are being used, and a Q&A session consisting of pre-planned and live audience discussion.
Here are three key takeaways from the event.
Applying Feist to Works
When determining whether or not a work has sufficient human authorship, the USCO drove home the reliance on the U.S. Supreme Court’s decision in Feist v. Rural Telephone, which established the de minimis test for human authorship. “As a constitutional matter,” the Court noted, “copyright protects only those constituent elements of a work that possess more than a de minimis quantum of creativity.”
“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,” the USCO further advises in Section 313.4(B) of the Compendium of U.S. Copyright Office Practices, Third Edition, (Compendium).
On March 16, the USCO released Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence to clarify its practices for reviewing copyright registration applications for works that include, whether entirely or partially, works generated by AI technologies. The policy guidance continues to build upon previously established legal principles and places the obligation on applicants in stating, “AI-generated content that is more than de minimis should be explicitly excluded from the application.”
During the event, the USCO’s Deputy Director of Registration Policy Erik Bertin explained that an applicant’s focus should be on the distinction between appreciable content and de minimis content. Bertin noted that disclosure is necessary for AI-generated content that doesn’t cross the de minimis threshold.
Based on some of the examples given, it would seem that the USCO would permit the registration of a sound recording released by The Beatles that utilized AI-powered mastering techniques. However, the AI-generated portions of television opening credits would require some level of disclaiming when registering.
Interestingly, the USCO discussed the use of generative AI to translate a work into a different language. They noted that copyright protections would not extend to the translation; however, an author should still register the work in its original language to afford protections that would extend to unauthorized uses of subsequent translations.
A number of the comments received by the Office focused on the question of whether prompts used in connection with generative AI tools are copyrightable. Rob Kasunic, Associate Register of Copyrights and Director of Registration Policy and Practice, shared that the USCO isn’t yet aware of any registrations of prompts. He added that they don’t prohibit anyone from submitting an application, so it is technically possible.
Bertin added that if a prompt has sufficiently protectable text or image content, it may be registrable. But he clarified that registration doesn’t necessarily extend protection to the output or other similar prompts or outputs.
When In Doubt, Disclose
The USCO explained that disclosure of non-human authored materials is not a new requirement. “Disclosures have always been required if the work contains unclaimable material,” they explained on a slide, noting the requirements to disclaim previously published material, previously registered material, public domain material, and copyrightable material owned by another party.
At multiple times throughout the event, the USCO reiterated that registration applications don’t have to include nuanced detail about where and how AI tools were used. One slide noted, “Checking a box or providing a brief, general statement is usually sufficient.” They also noted that excluded material doesn’t need to be removed from the deposit.
It’s important to remember that excluded materials are not protectable. The updated policy guidance and additional examples shared during the event do not change the USCO’s previous decision that generative works are not subject to copyright protection and therefore would be considered public domain materials.
Remember March 16
During the Q&A session, a pre-planned question from the USCO centered on the issue of an applicant not properly disclosing the use of AI. As previously reported on IPWatchdog, filing an application for registration with the USCO is subject to fines under 17 U.S.C. §506(e) for anyone that “knowingly makes a false representation of a material fact.”
In its response during the event, the USCO appeared to recognize the impact such a violation would have on applicants. They acknowledged that applications for registration submitted prior to the March 16 policy won’t be held to the higher standard to which post-March 16 applications for registration would be subject. This appears to be a direct reference to Section 4 of the policy guidance.
“Applicants who fail to update the public record after obtaining a registration for material generated by AI risk losing the benefits of the registration,” the USCO notes in the written policy guidance.
During the live stream, the USCO referenced the impact incorrect applications and subsequent registrations may have on litigation. Ultimately, an applicant’s failure to disclose within an application (or correct a previously issued registration) could allow a court to disregard a registration.
Other questions the Office took included whether the USCO is actively going back to prior registrations where AI played a role to initiate cancellations. The Office representatives confirmed they generally are not, and will defer to the courts on challenged registrations. The Office is also not requiring that applicants file a supplementary registration for such works, saying that’s a decision for the applicant to make. Those who wish to update pending registrations based on the new guidance should contact the USCO’s Public Information Office, which will help to clarify the scope of the claim.
The USCO noted that it cannot wait for courts to weigh in on these issues and that authors and claimants are waiting on decisions. In the meantime, the USCO remains in a legal battle with Dr. Stephen Thaler over its refusal to register a generative work that, based on the USCO’s most recent guidance, would fail to cross over the de minimis authorship threshold.
There’s a need for guidance and insight when it comes to the use of AI and ML technologies, including generative AI technologies, as evidenced by the more than 150 audience questions received during the live stream. The Office also noted that 1,500 people attended the virtual meeting. While the event focused exclusively on registration and protection of the output of generative AI tools, it’s important to remember there are additional issues along the entire lifecycle of generative AI.
At the end of the event, the USCO announced its next public webinar, “International Copyright Issues and Artificial Intelligence,” scheduled for July 26, 2023. The discussion will cover not only authorship but also training and infringement. Additionally, the event will feature leading international experts, including Columbia Law School’s Jane Ginsburg, the University of Sussex’s Andres Guadamuz, the University of Amsterdam’s Bernt Hugenholtz, and KU Leuven’s Luca Schirru, and the Universitat Oberta de Catalunya’s Raquel Xalabarder Plantada.