In recent months, generative artificial intelligence (AI) has raised numerous questions across various aspects of law and our society: have AI systems respected third-party rights during their development? Who owns the AI-generated content? Who may be liable for any infringement caused by AI-generated content? Concerning intellectual property (IP), the most prominent query is certainly whether AI-generated content can be protected by copyright or other IP rights and, if so, who holds the rights to it.
Initial answers to this intricate question have been provided in the United States and China. In the United States, both the US Copyright Office (“USCO”) and the Colombia District Court have denied copyright protection to AI-generated content, as well as to content created with the assistance of an AI tool1, due to the absence of significant human intervention. Conversely, in China, a pioneering decision ruled by the Beijing Internet Court granted copyright protection to an artwork generated by AI, considering AI as a “simple tool” at human’s disposal to generate creative content.
While the Chinese decision adds fuel to the already heated discussions concerning AI-generated content protection by copyright, it does not appear to contradict the reasoning behind the American decisions or the copyright principles in general. In fact, what underlies both American and Chinese approaches is the degree of human intervention in the AI-content creation.
Firstly, it is important to recall that only content resulting from human creativity can be protected by copyright in the U.S.2. The “human authorship requirement” is precisely what underlies the USCO’s analysis of the copyrightability of content generated by AI or created with AI assistance3.
In this sense, the USCO initially verifies if the work is the result of human authorship, in which case the AI system would only serve as a supporting tool, or if the traditional elements of authorship in the work were actually conceived and executed by an AI technology rather than a human. If the work contains AI-generated material, the USCO then verifies if the AI contributions result from a “mechanical reproduction” or from an author’s “own original mental conception, to which he gave visible form”4.
In other words, if the content is created autonomously by AI, without human creative intervention, it will not be protected by copyright. Conversely, if the AI tool user selects, arranges, or modifies AI-generated content in a sufficiently creative way (according to the specifications of the case), the resulting content may constitute an original work protected by copyright.
The USCO analysis is necessarily casuistic, its outcome depending on the circumstances of each case, especially the functioning of the AI tool and its role in creating the final work. However, as of now, the USCO has not agreed to grant copyright registration for AI-generated content. Four documented cases mark USCO’s refusal to register generative AI Work:
In February 2022, copyright protection was denied to the image “A Recent Entrance to Paradise”, which was entirely generated by the AI tool “Creativity Machine”. The USCO considered that Steven Thaler failed to demonstrate a human creative contribution when seeking copyright registration5. The U.S. District Court for the District of Columbia later upheld the USCO decision, noting that “copyright has never stretched so far as to protect works generated by new forms of technology operating absent any guiding human hand”.
In February 2023, the USCO partially revoked a copyright registration previously granted to Kristina Kashtanova for the graphic novel “Zarya of the Dawn”, which was created with the assistance of the AI MidJourney. After seeking clarification from the creator about her use of image-generating AI software, USCO considered Kashtanova as the author of the work’s text, as well as the selection, coordination, and arrangement of its written and visual elements, all protected by copyright. However, the images generated by Midjourney were not eligible for copyright protection, as they are not the product of human authorship.
This painting was created by MidJourney AI, as a result of a complex process of elaboration conducted by Jason Allen: 80 hours of work and thousands of prompts6, generating over 900 images. From these, he selected 3 images, which he later edited and printed on canvas. Allen tried to argue that MidJourney was used just a tool, “much like a brush”, creativity impulse being necessary to activate it and this painting won the annual fine arts competition at the Colorado State Fair in September 2022, USCO denied copyright registration, stating that Jason Allen’s sole contribution to MidJourney's image was to input the prompt and make visual modifications, not generating any traditional elements of authorship in the resulting output.
On December 11, 2023, the USCO rejected copyright registration for a two-dimensional computer-generated image titled “SURYAST”. This image was created from Ankit Sahni’s original photograph of a sunset, incorporating elements inspired by Vincent van Gogh’s “The Starry Night”, using a custom-developed software program called RAGHAV. Despite Sahni’s assertion that RAGHAV serves merely an “assistive software tool”, with creative control resting in his hands, the USCO concluded that “RAGHAV’s interpretation of Mr. Sahni’s photograph in the style of another painting is a function of how the model works and the images on which it was trained on—not specific contributions or instructions received from Mr. Sahni”.
Despite the USCO’s reiterated denial of granting copyright protection to AI-generated content as above, this possibility would not entirely be excluded. The USCO suggests that copyrightability for AI-generated content may be feasible if a significant human contribution can be discerned. The remaining question, however, is the extent of human creativity required by the USCO for AI-generated content to be eligible for copyright protection.
On November 27, 2023, the Beijing Internet Court rendered a historical decision by granting copyright protection to an AI-generated image in the context of a copyright infringement action. In this case, an individual named Li created the image of a young woman through Stability Diffusion7 and published it on a Chinese lifestyle platform called “Little Red Book”. Subsequently, a blogger named Liu used the image to illustrate a poem published on the content-sharing platform Baijiahao without the Plaintiff’s previous authorization, leading him to file a suit for copyright infringement. In response, Defendant contested Plaintiff’s ownership rights to the image, compelling him to demonstrate the creative process and the claimed copyrights.
Arguing that the AI systems are tools for humans to create expressive works, the Plaintiff drew an analogy to the creative process of a photograph. The “intellectual investment” of selecting and arranging the inputs necessary to generate the image would be similar to a photographer’s manual adjustments with antique cameras to achieve the desired result in a photograph. In this sense, the individual creating expressive works through AI systems is posited as the author, just as the photographer is considered the author of a copyright-protected photograph.
Additionally, the plaintiff presented evidence of a meticulous process of creation, involving the selection of prompts (both positive and negative), as well as the selection of outputs, which were later modified to narrow down the output8, concerning both the personage and the landscape depicted in the image.
Considering this, and stating that an AI-generated image will be eligible for copyright protection if it reflects the original intellectual investment of a human being, the Beijing Internet Court determined that a sufficient amount of intellectual achievement had been reached in the case and, thus, copyright protection should be granted. In other words, the court considered that the prompts and modifications effected by the Plaintiff reflect his aesthetic choices and personality judgment, demonstrating his active and predominant role in the creation of the image. Therefore, copyright protection was granted to the image generated by Stable Diffusion, and the Defendant was held liable for copyright infringement.
The decision of the Beijing Internet Court decision was deemed “tech-friendly”, aligning with the Chinese government’s policy of providing financial and political support for the development of its domestic AI industry. Nevertheless, the decision is still subject to appeal before the Beijing Intellectual Property Court and may still be reversed.
Despite the divergent outcomes in the American and Chinese decisions, they share a common foundation – the need for human intervention for AI content to be eligible for copyright protection. However, the criteria adopted by the USCO when analyzing human intervention in the process of creation seem to be more stringent than those adopted by the Chinese court. If the USCO denied or limited the scope of copyright protection to AI-generated content, it was precisely because it considered they lacked human authorship.
The case “Theatre D’Opéra Spatial” notably illustrates that the degree of significant human creative intervention in AI content required by the USCO might be a real obstacle to copyright protection. Despite the complexity of the creation process for this image, which involved as much human intervention as the case submitted to the Beijing Internet Court, the USCO rejected the applicant’s registration request, considering that it lacked traditional elements of authorship in the resulting output.
Nonetheless, as the USCO’s position does not exclude the possibility of granting copyright protection for AI content and the analysis is necessarily casuistic, we wonder if the Beijing Internet Court’s decision may have a ripple effect on future similar rulings in the U.S. and Worldwide, especially considering that the Chinese ruling presents arguments that could easily align with other copyright legislations, including French and European legislations.
Indeed, in the current uncertainty in the AI legal landscape and the growing cases concerning copyright protection of AI-generated content, the Chinese decision might exert significant influence not only on other administrative or judicial decisions worldwide but also on the development of AI itself, especially if subsequently confirmed by a superior court.
1 While AI-generated content is created entirely by an AI algorithm, and AI-assisted content is created with some level of human input, the distinction between them is becoming increasingly blurred as AI algorithms become more sophisticated.
2 In the U.S., copyright protection is effective upon the creation of the work. However, for legal opposition in the courts, a work must be registered beforehand with the U.S. Copyright Office.
3 The U.S. Copyright Office (USCO) is a Federal agency in the U.S. tasked with administering the copyright registration system, as well as advising Congress and the Federal Judiciary on copyright and related matters.
4 The U.S. Copyright Office published in Mars 2023 its Guidelines on AI: “Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence”.
5 U.S. Copyright Off., Re: Second Request for Reconsideration for Refusal to Register A Recent Entrance to Paradise, Feb. 14, 2022
6 Prompts are textual descriptions provided by the AI tool user.
7 Stable Diffusion is a “text-to-image” AI program launched by Stability AI Ltd. In August 2022.
8 The prompts used by the Plaintiff: “ultra-photorealistic: 1.3), extremely high quality highdetail RAW color photo, in locations, Japan idol, highly detailed symmetrical attractive face, angular symmetrical face, perfect skin, skin pores, dreamy black eyes, reddish-brown plaits hairs, uniform, long legs, thighhighs, soft focus, (film grain, vivid colors, Film emulation, kodak gold portra 100, 35mm, canon50 f1,2), Lens Flare, Golden Hour, HD, Cinematic, Beautiful Dynamic Lighting”.