On 16 March 2023, the US Copyright Office (“USCO”) published guidance on the registration of works containing AI-generated content. The USCO’s policy statement was released against the backdrop of the proliferation of generative AI tools which are able to create content based on user prompts. The USCO ultimately concluded that the “authorship” requirement of US copyright law refers to “human authorship” (in line with prior case law) and appears to reject the extension of copyright to works generated with the aid of AI technology outside of the user’s control.

The USCO has had to consider the application of established copyright law principles to AI-generated works in light of an increase in applications claiming copyright in such works. For example, the USCO granted the first copyright certificate for an AI-generated work  in September 2022 for ‘Zarya of the Dawn’, a graphic novel which was created by Kristina Kashtanova, who used Midjourney technology (an AI program) to generate the graphics in the novel based on the text. In February 2023, the USCO cancelled the original certificate issued to Kashtanova because the images in the work that were generated by the Midjourney technology “are not the product of human authorship” and because the registration for the work did not “disclaim its Midjourney-generated content”. A new copyright certificate granting protection only to the human-authored elements was subsequently issued by the USCO.

The USCO’s guidance clarifies the general approach the Office will adopt with respect to the registration of works containing AI-generated content in line with the principles applied in the specific ‘Zarya of the Dawn’ matter. In essence, the USCO has clarified that copyright protection will only apply to contributions which reflect a human’s “own mental conception” and not to contributions which result from “mechanical reproduction”. However, under the USCO’s guidance, whether an AI-generated work will be entitled to copyright protection is heavily fact-specific, specifically with regard to “how the AI tool operates and how it was used to create the final work”. Below, we set out our key takeaways from the guidance.

Human authorship  The Guidance reaffirms that an “author” must be human and that copyright functions to protect material that is the product of human creativity: “to qualify as a work of ‘authorship’ a work must be created by a human being”. 

The Guidance clarifies that a case-by-case analysis is required to determine if a work containing AI-generated material is the result of only “mechanical reproduction” or if it involves “own original mental conception”.

If the work’s “traditional elements of authorship were produced by a machine”, for example if the only human interference is the insertion of a text prompt and the AI generates an image, the work lacks human authorship and “creative control” by the human user. As such, the USCO will not entitle the portion created by the AI to copyright protection.

The USCO clearly stated that mere “influence” by a human over the AI input or output is not sufficient to render the work registrable.

The USCO distinguished tools like AI from traditional digital tools used by authors via the notion that the authors had no way to control or predict the outcome of AI-generated works, as opposed to such traditional tools which are fully under the control of the author (e.g., edits made to a photo in Adobe’s Photoshop).
Manipulation of the outputWhere there is a significant level of human editing of AI-generated content – via selection, arrangement, coordination, or other editing – the human-added elements may entitled to copyright protection. However:

  such protection is only available where the level of human authorship is sufficient to support a claim as to the copyrightability of such content, and

  such copyright protection would only be available for the elements the human added, not the underlying AI-generated basis. In other words, the copyright would apply only to the human-authored aspects that are independent of the AI-generated material.
Copyrightability of the inputThe level of detail and creativity of some prompts (the “inputs” to the AI system) may render such prompts themselves protectable by copyright. However, this does not mean that the “material generated from a copyrightable prompt is itself copyrightable.”
Applications for registrationApplicants must disclose that the works applied for contain AI-generated content via the Standard Application:

  the human author(s) should be identified in the “Author Created” field and a brief statement should be provided. By way of example, an applicant who incorporates AI-generated text into a larger work should claim the portions of the text that the human authored. Applicants should not list an AI tool as an author or co-author; and

  in the “Limitation of Claim” section of the application, applicants must describe and disclaim any AI-generated content that is not de minimis.

Pending applications that do not follow the guidelines should be corrected by either contacting the USCO or by submitting a supplementary registration application.

Existing registrations should be corrected by the applicant if non-de minimis AI-generated content was not disclosed. If the public record is not corrected, the USCO may cancel the registration, and/or a court may disregard such a registration in an infringement action.
More to come…The USCO states that it “recognizes that AI-generated works implicate other copyright issues not addressed in this statement.” An agency-wide initiative has been launched to look into the various other issues that arise in the context of AI-generated material.

The USCO also announced that it intends to publish a notice of inquiry later this year “seeking public input on additional legal and policy topics, including how the law should apply to the use of copyrighted works in AI training and the resulting treatment of outputs”.

It is clear that the USCO is monitoring factual and legal developments in this area and additional guidance relating to this new technology should be expected.